What this PR does

Before this PR:

• When reasoning_effort was set to "default", temperature and topP parameters were incorrectly suppressed because "default" is a truthy string value.
• Claude 3.7/4.0/4.5 models had no dedicated thinking model type, falling back to "default" which gave them generic reasoning effort options. Users could not select "none" to disable thinking even though the Anthropic API supports it.
• TopP was fully disabled for Claude reasoning models with reasoning effort set, instead of being clamped to a valid range.
• Parameter adjustments (e.g., temperature clamping, topP disabling) happened silently with no user-visible feedback.

After this PR:

• Explicitly checks reasoning_effort !== 'default' and reasoning_effort !== 'none' so that temperature and topP are only disabled for actual reasoning effort values.
• Adds a 'claude' thinking model type for Claude 3.7/4.0/4.5 that supports 'none' but not 'xhigh' (options: ['default', 'none', 'low', 'medium', 'high']).
• TopP is clamped to [0.95, 1] for Claude reasoning models with thinking enabled, instead of being fully disabled.
• Refactors getTemperature/getTopP by inlining helper functions and moving enable-checks to the top as early returns.
• Adds logger.info messages so users can see when parameters are adjusted or disabled in the log (e.g., temperature clamped, topP dropped due to mutual exclusivity).

Fixes #13503

Why we need it and why it was done in this way

The following tradeoffs were made:

• Inlined getTemperatureValue/getTopPValue helper functions — they added indirection without value, and the enable-check was buried at the end instead of being a clear early return.
• Added logger.info for user-visible feedback on parameter adjustments — info level is visible in user logs, helping users understand why their settings were modified.
• Clamped topP to [0.95, 1] per Anthropic extended thinking docs rather than fully disabling it.
• Added a separate 'claude' thinking type instead of reusing 'claude46', since Claude 3.7/4.0/4.5 do not support xhigh reasoning effort.

The following alternatives were considered:

• Keeping the helper functions but fixing only the reasoning_effort check — rejected because the refactor makes the logic significantly easier to follow.
• Using a single 'claude' type for all Claude models — rejected because 4.6 supports xhigh which older models don't.

Breaking changes

None

Special notes for your reviewer

• The enableTemperature/enableTopP checks are moved to the top of each function as early returns, making the control flow clearer.
• Temperature now falls back to DEFAULT_ASSISTANT_SETTINGS.temperature and the isMaxTemperatureOneModel clamp always runs (previously skipped when temperature was 0).
• The Claude branch is placed first in _getThinkModelType for early matching.

Checklist

• PR: The PR description is expressive enough and will help future contributors
• Code: Write code that humans can understand and Keep it simple
• Refactor: You have left the code cleaner than you found it (Boy Scout Rule)
• Upgrade: Impact of this change on upgrade flows was considered and addressed if required
• Documentation: A user-guide update was considered and is present (link) or not required. Check this only when the PR introduces or changes a user-facing feature or behavior.
• Self-review: I have reviewed my own code (e.g., via /gh-pr-review, gh pr diff, or GitHub UI) before requesting review from others

Release note

Fix temperature and topP parameters being incorrectly disabled when reasoning effort is set to "default" or "none" for Claude reasoning models. Add "none" option to disable thinking for Claude 3.7/4.0/4.5 models. TopP is now clamped to [0.95, 1] when thinking is enabled instead of being fully disabled.