llm: group providers under llm/providers/ subpackage

Mirrors the openarmature.graph.middleware shape so the package layout
signals where future providers (Anthropic, Bedrock, gateway-shaped)
will live. Public surface is unchanged — `from openarmature.llm import
OpenAIProvider` still works via the package-root re-export.

Author: chris-colinsky

src/openarmature/llm/__init__.py

@@ -46,8 +46,8 @@
     ToolMessage,
     UserMessage,
 )
-from .openai import OpenAIProvider
 from .provider import Provider, validate_message_list, validate_tools
+from .providers import OpenAIProvider
 from .response import FinishReason, Response, RuntimeConfig, Usage
 
 __all__ = [

src/openarmature/llm/providers/__init__.py

@@ -0,0 +1,19 @@
+"""Concrete provider implementations.
+
+Each provider talks to a specific LLM backend's wire format and
+satisfies the :class:`openarmature.llm.Provider` Protocol. The current
+catalog ships :class:`OpenAIProvider` (OpenAI Chat Completions wire
+format — also covers vLLM, LM Studio, llama.cpp). Anthropic / Bedrock /
+gateway-shaped providers will land here in later phases.
+
+Users typically import from the package root::
+
+    from openarmature.llm import OpenAIProvider
+
+This subpackage exists to keep the provider catalog grouped and to
+mirror the ``openarmature.graph.middleware`` layout.
+"""
+
+from .openai import OpenAIProvider
+
+__all__ = ["OpenAIProvider"]

src/openarmature/llm/openai.py src/openarmature/llm/providers/openai.pysrc/openarmature/llm/openai.py renamed to src/openarmature/llm/providers/openai.py

@@ -44,7 +44,7 @@
 import httpx
 from pydantic import ValidationError
 
-from .errors import (
+from ..errors import (
     ProviderAuthentication,
     ProviderInvalidModel,
     ProviderInvalidRequest,
@@ -53,16 +53,16 @@
     ProviderRateLimit,
     ProviderUnavailable,
 )
-from .messages import (
+from ..messages import (
     AssistantMessage,
     Message,
     SystemMessage,
     Tool,
     ToolCall,
     UserMessage,
 )
-from .provider import validate_message_list, validate_tools
-from .response import FinishReason, Response, RuntimeConfig, Usage
+from ..provider import validate_message_list, validate_tools
+from ..response import FinishReason, Response, RuntimeConfig, Usage
 
 
 class OpenAIProvider: