Reasoning effort fix

[harshiv-26]

Note

Low Risk
Changes are catalog/metadata only (params and removeParams); no runtime code paths. Wrong values could mislead API clients but are easy to spot in config review.

Overview
This PR corrects how reasoning_effort is declared on provider model YAML so clients only see the knob where it applies, with per-model allowed values.

Cohere adds reasoning_effort to default.yaml (default high, none/high) for chat-capable configs, and strips it from embed, rerank, and transcribe models via removeParams.

AWS Bedrock Mantle GPT-OSS chat models now expose reasoning_effort with low/medium/high.

Google Vertex Anthropic chat entries are split: many Opus/Fable/Sonnet variants gain explicit reasoning_effort (defaults and tiers like xhigh/max where relevant), while Haiku 4.5 and Sonnet 4.5 entries remove reasoning_effort so it is not offered on those Vertex routes.

^{Reviewed by Cursor Bugbot for commit 888a72f. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

There are 2 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 21342ca. Configure here.}