[alexcastgonz (pcbeingused333)]

GPT-5 mini, GPT-5 nano and GPT-5 are retired by OpenAI on 2026-12-11 (replacements: gpt-5.6-terra, gpt-5.6-luna, gpt-5.6-sol). The SDK fallback model, get_default_model() in Python and getDefaultModel() in TypeScript, is still gpt-5-mini. After the retirement, any scorer that runs without init(default_model=...) or an explicit model would fail.

#230 moved the exported DEFAULT_MODEL constant to gpt-6-luna and kept this fallback on purpose. This PR changes the fallback now, because its model goes away:

• get_default_model() / getDefaultModel() return gpt-5.6-terra, OpenAI's replacement for gpt-5-mini. It keeps the gpt-5 prefix, so the routing to the Responses API does not change.
• The deprecated DEFAULT_RAGAS_MODEL (gpt-5-nano) becomes gpt-5.6-luna, its replacement.
• Docstrings and SCORERS.md name the new defaults.

If you prefer one default for both the constant and the fallback, gpt-6-luna is a two-string change; it does not start with gpt-5, so it would use the Chat Completions path unless use_responses_api is set. Say which one you prefer and the PR will follow it.

Tests: the Python tests for the fallback (test_oai.py, test_init_models.py) and the TypeScript ones (oai.test.ts, init-models.test.ts) now expect the new default; the four Python ones fail on main. Python: those two files pass. TypeScript: oai.test.ts, init-models.test.ts and llm.test.ts pass. The tests that call the live API (ragas.test.ts, the Python test_llm.py ones) need an OpenAI key and were not run, so the full suites were not compared against the previous state.

This change was prepared with AI assistance (Claude Code).