refactor!: remove in-app AI/LLM stack — RFlect is now zero-dependency (Workstream A)

RFlect predated MCP and made its own outbound LLM API calls for an in-app
chat assistant, GUI AI-report generation, and AI prose inside the MCP
generate_report tool. With an MCP agent driving the tools, the agent is the
LLM — so RFlect no longer needs to call one. This removes the entire
AI/LLM/API-key surface; the report path is deterministic and the optional
agent-authored-narrative interface lands in a follow-up (Workstream B).

BREAKING:
- Removed plot_antenna/llm_provider.py and plot_antenna/api_keys.py.
- generate_report no longer takes ai_* options or calls any LLM; it produces
  deterministic, data-driven summaries/conclusions/captions.
- No API key or paid subscription is required for any RFlect feature.

Removed
- plot_antenna/gui/ai_chat_mixin.py (in-app chat) + its mixin wiring.
- GUI Tools-menu AI items (Manage API Keys, AI Settings, Generate Report
  with AI, AI Chat Assistant) + their dialog methods in dialogs_mixin.
- The AI-vision RFAnalyzer internals in save.py; RFAnalyzer is now a
  deterministic caption/summary generator.
- _create_llm_provider / _generate_ai_text + ai_* ReportOptions fields in
  report_tools; provider plumbing in orchestration.
- AI_* keys from config_template; ai_analysis/ai_prompt/ai_generated flags
  from report templates.
- tests/test_llm_provider.py, tests/test_api_keys.py, and the LLM-specific
  test classes in test_mcp_report.

Renamed
- plot_antenna/ai_analysis.py -> analysis_engine.py (it was always pure
  numpy compute — gain stats, HPBW, sidelobe, polarization, horizon — and
  backs every MCP analysis tool; the old name was a misnomer). Importers and
  tests/test_ai_analysis.py -> test_analysis_engine.py updated.

Verification: full suite 313 passed / 148 skipped; MCP server + GUI import
with no openai/anthropic packages installed; zero residual AI/LLM/key refs.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: RFingAdam

plot_antenna/ai_analysis.py plot_antenna/analysis_engine.pyplot_antenna/ai_analysis.py renamed to plot_antenna/analysis_engine.py

@@ -151,108 +151,6 @@
 HEADING1_FONT_SIZE = 18  # Section heading font size
 HEADING2_FONT_SIZE = 14  # Subsection heading font size
 
-# ============================================================================
-# AI PROVIDER CONFIGURATION
-# ============================================================================
-
-# AI Provider: "openai", "anthropic", or "ollama"
-AI_PROVIDER = "openai"
-
-# Provider-specific model defaults
-AI_OPENAI_MODEL = "gpt-4o-mini"
-AI_ANTHROPIC_MODEL = "claude-sonnet-4-20250514"
-AI_OLLAMA_MODEL = "llama3.1"
-AI_OLLAMA_URL = "http://localhost:11434"
-
-# Vision model override (for plot analysis)
-# Set to a vision-capable model if your default model doesn't support vision
-AI_VISION_MODEL = None  # None = use same as AI_MODEL. e.g. "llava" for Ollama
-
-# ============================================================================
-# AI ANALYSIS CONFIGURATION (OpenAI)
-# ============================================================================
-
-# AI Model Settings
-AI_MODEL = "gpt-4o-mini"  # OpenAI model to use (default: gpt-4o-mini for compatibility)
-#
-# RECOMMENDED MODELS FOR AI CHAT (with function calling):
-#
-# GPT-5.2 FAMILY (LATEST - Uses Responses API):
-# - "gpt-5.2" - Flagship model, best quality, supports tools + vision
-# - "gpt-5-mini" ⭐ RECOMMENDED - Great balance of speed/cost/quality
-# - "gpt-5-nano" - Fastest, lowest cost, good for simple analysis
-#
-# GPT-4 FAMILY (LEGACY - Uses Chat Completions API):
-# - "gpt-4o-mini" - Best legacy value, fast, supports functions + vision
-# - "gpt-4o" - Premium quality, supports functions + vision
-# - "chatgpt-4o-latest" - Latest 4o version
-#
-# ❌ NOT COMPATIBLE WITH AI CHAT:
-# - "o1-preview", "o1-mini", "o3-mini" - NO function calling support
-#   These models cannot call Python functions interactively
-#
-# GPT-5.2 ADVANTAGES:
-# ✅ Uses new Responses API with reasoning control
-# ✅ Supports tools (function calling) with strict schemas
-# ✅ Supports vision (image analysis for plots)
-# ✅ Configurable reasoning effort (none/low/medium/high)
-# ✅ Configurable text verbosity (low/medium/high)
-# ✅ Better structured output support
-#
-# WHEN TO USE ALTERNATIVES:
-# - Use "gpt-5.2" for best quality analysis
-# - Use "gpt-5-mini" for balanced speed/quality
-# - Use "gpt-4o-mini" if GPT-5 API access not available
-# - DON'T use O-series models for AI chat (no function calling)
-AI_MAX_TOKENS = 150  # Maximum response length for report captions (150 = ~100 words)
-AI_CHAT_MAX_TOKENS = 800  # Maximum response length for AI Chat assistant (800 = ~500 words)
-# NOTE: For GPT-5, this maps to verbosity levels:
-# ≤100 → "low", ≤250 → "medium", >250 → "high"
-AI_TEMPERATURE = 0.2  # Lower = more consistent (0.0-1.0)
-# NOTE: GPT-4 uses this directly. GPT-5 maps it to reasoning.effort:
-# ≤0.1 → "none", ≤0.3 → "low", ≤0.6 → "medium", >0.6 → "high"
-AI_TOP_P = 0.8  # Nucleus sampling parameter (GPT-4 only, not used for GPT-5)
-AI_TIMEOUT_SECONDS = 30  # API request timeout
-
-# AI Prompt Settings
-AI_RESPONSE_STYLE = "concise"  # "concise" or "detailed"
-# - "concise": Single-paragraph technical descriptions (~80 words)
-# - "detailed": Multi-section analysis with recommendations (~200+ words)
-AI_INCLUDE_RECOMMENDATIONS = False  # Include design recommendations in each caption
-# Set True for educational reports, False for concise documentation
-AI_MAX_WORDS = 80  # Target word count for AI responses
-# Suggested ranges: 60-80 (concise), 150-250 (detailed), 300+ (research-level)
-
-# Advanced AI Parameters (GPT-5.2 Responses API)
-# These parameters are ONLY used when AI_MODEL is a GPT-5 variant (gpt-5*, gpt-5.2)
-# For GPT-4 models, these are ignored and Chat Completions API parameters (temperature, top_p) are used instead
-
-AI_REASONING_EFFORT = "low"  # GPT-5.2 reasoning intensity: "none", "low", "medium", "high", "xhigh"
-# - "none": No reasoning tokens, fastest response (like GPT-4 behavior)
-# - "low": Light reasoning for straightforward technical analysis (RECOMMENDED for RF reports)
-# - "medium": Balanced reasoning for complex patterns
-# - "high": Deep reasoning for research-level analysis (slower, more expensive)
-# - "xhigh": Maximum reasoning effort for most complex problems
-# NOTE: RF antenna analysis is domain-specific but not highly complex reasoning,
-# so "low" provides best speed/quality/cost balance
-
-AI_TEXT_VERBOSITY = "auto"  # GPT-5.2 output verbosity: "auto", "low", "medium", "high"
-# - "auto": Let model decide based on AI_MAX_TOKENS (RECOMMENDED)
-# - "low": Terse, minimal prose (good for captions)
-# - "medium": Balanced detail (good for technical reports)
-# - "high": Comprehensive explanations (research papers)
-# When "auto", maps AI_MAX_TOKENS: ≤100→low, ≤250→medium, >250→high
-
-AI_GENERATE_REASONING_SUMMARY = False  # GPT-5.2 only: Include reasoning summary in response
-# When True, provides insight into model's thought process
-# Useful for debugging or understanding complex analysis decisions
-# Adds ~50-100 tokens to response, minimal cost impact
-
-AI_USE_STRUCTURED_OUTPUT = False  # Enable JSON structured output for consistent formatting
-# When True, enforces schema-based responses (requires schema definition in code)
-# Useful for programmatic parsing but reduces natural language quality
-# RECOMMENDED: Keep False for readable report captions
-
 # ============================================================================
 # LINK BUDGET / RANGE ESTIMATION
 # ============================================================================

plot_antenna/api_keys.py

@@ -7,8 +7,7 @@
 Structure:
 - main_window.py: Core AntennaPlotGUI class combining all mixins
 - base_protocol.py: Protocol/interface definitions for type checking
-- dialogs_mixin.py: Dialog methods (About, API Key, AI Settings, Settings)
-- ai_chat_mixin.py: AI chat window and all AI-related methods
+- dialogs_mixin.py: Dialog methods (About, Settings)
 - tools_mixin.py: Bulk processing, polarization analysis, converters
 - callbacks_mixin.py: File import, data processing, save operations
 - utils.py: Shared utility functions (DualOutput, resource_path, etc.)