feat: configurable embedding provider, model, and dimensions via environment variables

[Ashwin-3cS]

Description

The server currently hardcodes openai/text-embedding-3-small for embeddings and openai/gpt-4o-mini for LLM calls. This limits flexibility for self-hosters who may want to:

• Use a dedicated embedding provider (Jina, Cohere, etc.) with a separate key or budget
• Use OpenRouter free models instead of OpenAI directly
• Customize embedding dimensions (e.g. 1024 for Jina v3)

At the moment, achieving this requires modifying the source code.

---

Changes

Configuration (types.rs)

Adds five new Config fields, all read from environment variables:

Variable	Default	Purpose
EMBEDDING_API_KEY	falls back to OPENAI_API_KEY	API key for embedding provider
EMBEDDING_API_BASE	falls back to OPENAI_API_BASE	Base URL for embedding provider
EMBEDDING_MODEL	openai/text-embedding-3-small	Embedding model identifier
EMBEDDING_DIMENSIONS	omitted	Optional output dimension override (e.g. 1024)
LLM_MODEL	openai/gpt-4o-mini	LLM used for /api/analyze and /api/ask

This improves flexibility for developers in the ecosystem by allowing different providers, models, and configurations without modifying source code.

---

Embedding + LLM usage (routes.rs)

• generate_embedding() now uses a fallback chain:
  • EMBEDDING_API_KEY → OPENAI_API_KEY
  • EMBEDDING_API_BASE → OPENAI_API_BASE
• Adds optional dimensions field to embedding API request
  • Uses skip_serializing_if = "Option::is_none" for provider compatibility
• Replaces hardcoded LLM model with config.llm_model in all call sites
• Mock embedding path respects EMBEDDING_DIMENSIONS for consistency in dev/test

---

Server startup (main.rs)

• Logs active configuration at startup:
  • embedding model
  • base URL
  • optional dimensions
  • LLM model
• Emits a WARN if EMBEDDING_DIMENSIONS does not match the schema column dimension
  • Prevents silent breakage in cosine similarity queries

---

Environment config (.env.example)

• Documents all new environment variables
• Includes commented example values (e.g. Jina setup)

---

Documentation (docs/)

• Updated:
  • environment-variables.md
  • self-hosting.md
• Includes:
  • Explanation of fallback chains
  • Guidance on using custom providers
  • Warning against mixing embedding dimensions mid-deployment

Documentation has been updated to reflect these changes. If any adjustments or clarifications are needed, feel free to point them out and I'll follow up in this PR.

---

Backwards Compatibility

All new variables have defaults that preserve the current behavior.
Existing deployments with no .env changes remain unaffected.

---

Test

• Verified locally with:
  • Jina embeddings (EMBEDDING_MODEL=jina-embeddings-v3, EMBEDDING_DIMENSIONS=1024)
  • OpenRouter free LLM
• Server startup logs correctly reflect active configuration
• Dimension mismatch warning triggers when schema and config differ