* feat(snowflake-cortex): add Claude Opus 5, Sonnet 5, Opus 4.6 and Opus 4.5
* fix(snowflake-cortex): align Claude reasoning_options with tested chat-completions surface
Verified against POST /api/v2/cortex/v1/chat/completions:
- Opus 5 / Sonnet 5: reasoning.effort and reasoning.max_tokens return 400.
reasoning_effort, output_config.effort and thinking.type return 200 but are
ignored (reasoning_effort=bogus_zzz also returns 200) and never produce
reasoning_details, so no caller control is exposed -> [].
- Opus 4.6 / 4.5: reasoning.max_tokens is the only field that actually engages
thinking (sole case returning reasoning_details) -> budget_tokens. Effort
values are not read (effort=bogus_zzz behaves identically), and max_tokens=100
is accepted, so no effort enum and no min bound.
rognit
0542ac135a
