feat(embeddings): OpenAI-compatible local endpoint support

[jonathanhfmills]

Summary

• Adds LUCID_EMBEDDING_URL and LUCID_EMBEDDING_MODEL env vars for keyless local embedding servers (Ollama, LM Studio, vLLM, etc.)
• Custom endpoint checked first in provider detection — highest priority
• API key optional for local endpoints; isAvailable() probes /models to verify before committing
• Enables long-context models (e.g. nomic-embed-text at 8192 tokens) as an alternative to the built-in BGE model (512-token limit)

Usage

LUCID_EMBEDDING_URL=http://localhost:11434/v1 \
LUCID_EMBEDDING_MODEL=nomic-embed-text \
lucid-server

Provider detection order: LUCID_EMBEDDING_URL → local ONNX (BGE) → OpenAI → none.

Changes

• EmbeddingConfig: new openaiBaseUrl field
• embedOpenAIBatch: uses configurable baseUrl, API key optional
• isAvailable: probes /models when openaiBaseUrl set
• detectProvider: checks LUCID_EMBEDDING_URL first

Also includes

• AGENTS.md docs — 24 hierarchical context files (one per directory) providing structured AI-readable documentation: purpose, key files, conventions, testing requirements, and dependency relationships. CLAUDE.md and GEMINI.md symlink to each AGENTS.md so Claude, Gemini, and other AI agents load the right context automatically when working in any subdirectory.
• CHANGELOG filled in for 0.6.1–0.6.5 (entries were missing for all point releases since the 0.6.0 audit).

Test plan

• LUCID_EMBEDDING_URL=http://localhost:11434/v1 LUCID_EMBEDDING_MODEL=nomic-embed-text lucid-server — verify embeddings work via Ollama
• lucid status — verify custom endpoint shown as active provider
• No env vars set — verify fallback to BGE still works
• Invalid URL — verify graceful failure and lucid status shows skip reason
• bun test passes

[jonathanhfmills]

PR #8 only touches install.sh, our PR touches embeddings.ts, cli.ts, CHANGELOG.md, and AGENTS.md files. Zero conflicts.

And PR #8 actually solves the exact pain we hit today — having to manually cp cli.ts ~/.lucid/server/src/cli.ts. Once that merges, ./install.sh --local from the repo root handles it 👍

Worth merging PR #8 first, then this one lands clean on top.

I trust your judgment, @JasonDocton. Hope all's well on your end! LMK if you need anything from me 🚀