fix: reasoning_effort "max" silently became "medium" for Claude

[Fwind43]

What

Add 'max': 'max' to the Claude reasoning_effort mapping in _apply_claude_thinking (llmcore.py).

Why

reasoning_effort accepts max at config level (_enum includes it), but the Claude mapping only knew low/medium/high/xhigh. Setting max hit the WARN branch and dropped the field entirely; then the fake_cc_system_prompt path injected the default {"effort": "medium"} — so a user who configured max silently sent medium to the API.

xhigh -> max still works unchanged; max is now a direct alias.

Testing

• py_compile passes.
• Unit-exercised _apply_claude_thinking on a stub: max -> max, xhigh -> max, medium -> medium, none -> WARN/ignored (unchanged).