feat: add text_completion, generate_embeddings and create_response tools

[infinitimeless]

Summary

Adds three new MCP tools to the bridge, bringing the total from 4 to 7.

Credit

These tools were originally proposed and implemented by @ahmedibrahim085 in PR #5. That PR was closed after a long delay on our side, during which Ahmed went on to build the excellent lmstudio-bridge-enhanced. This implementation is a clean re-write inspired by his work, with the issues identified during code review addressed.

New Tools

text_completion

Raw text/code completion via /v1/completions. Simpler and faster than chat_completion for single-turn tasks like code continuation — no chat formatting overhead.

generate_embeddings

Generate vector embeddings via /v1/embeddings. Supports single text or batch input. Foundation for RAG and semantic search workflows.

create_response

Stateful conversations via /v1/responses. Maintains context automatically via previous_response_id — no manual message history management needed. Requires LM Studio v0.3.29+.

Changes vs PR #5

All issues identified in the code review have been addressed:

• No bare except — model auto-detection failure returns a clear JSON error asking the user to specify a model explicitly
• No hardcoded model name — removed the "qwen/qwen3-coder-30b" fallback entirely
• Malformed JSON guard — generate_embeddings validates the response is parseable before returning it
• Server name unchanged — still "lmstudio-bridge", no breaking change for existing users
• Port validation — LMSTUDIO_PORT is validated as an integer in range 1–65535
• Timeout on all requests — chat_completion now also has timeout=60
• No extraneous files — only lmstudio_bridge.py is changed