Support custom base URL for major providers - #196

Summary

• Adds ANTHROPIC_BASE_URL, OPENAI_BASE_URL, and GOOGLE_BASE_URL config support so all three chat providers can be routed through a custom endpoint (e.g. an internal LLM gateway), each forwarded to its LangChain client's base_url param when set.
• Fixes Gemini specifically: langchain-google-genai 3.1.0's underlying GAPIC client defaults to gRPC, which requires a bare hostname:port and silently hangs against a path-based REST gateway URL; forcing transport="rest" as a workaround then hit a second bug where that version's async client has no real async REST transport ('GenerateContentResponse' object can't be awaited). Resolved by upgrading langchain-google-genai to 4.3.4, which moves to Google's REST/async-native google-genai SDK — no transport workaround needed.
• Bumps langchain-anthropic (1.4.7 → 1.5.6) and langchain-openai (1.0.3 → 1.5.1, pulling openai SDK 2.8.1 → 3.0.0) for general dependency hygiene while in here — these two already worked correctly with a custom base URL since both providers are REST-only.

Test plan

• uv run pytest -m "not live" — 981 passed, no regressions
• Live conversation generation through the gateway for Gemini (gemini-3.5-flash), Claude (claude-sonnet-5), and OpenAI (gpt-5.2) via generate.py — all three completed with real, coherent transcript output
• New unit tests covering *_BASE_URL config loading and forwarding to each provider's LangChain client (test_config.py, test_claude_llm.py, test_gemini_llm.py)