fix(core): send Grok reasoning effort on Bedrock Converse - #51151

Issue for this PR

N/A. Found while testing default output limits for #51021.

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Grok 4.6's reasoning effort variants on Bedrock Converse (us.xai.grok-4.6, global.xai.grok-4.6) now send the field Bedrock actually reads.

They used to fall into the generic branch of variant.ts and send Nova's additionalModelRequestFields.reasoningConfig.maxReasoningEffort. Bedrock silently ignores unknown fields for Grok, so every request ran at Grok's default effort (low on Bedrock).

They now send additionalModelRequestFields.reasoning.effort, the same shape we already use for OpenAI models on Bedrock. AWS's Grok 4.6 model card documents effort through the reasoning parameter.

• How I confirmed the field: I sent an invalid effort through each candidate field. Only a field Bedrock reads rejects it.

  Field	"bogus" effort
  reasoning: { effort }	400: "Supported values are: 'none', 'low', 'medium', 'high', 'xhigh', and 'max'"
  reasoning_effort	200 (ignored)
  reasoningConfig.maxReasoningEffort (before)	200 (ignored)

• Other Grok routes: xAI direct and Bedrock Mantle already use the Responses API with reasoning.effort, and Mantle rejects an invalid effort the same way.

• Nova 2: still gets reasoningConfig.

How did you verify your code works?

• Unit test: covers Bedrock Converse effort variants for Grok 4.6 and Nova 2. The packages/core variant and models-dev suites pass.
• Live against Bedrock: Grok 4.6 through this branch sent additionalModelRequestFields.reasoning.effort at xhigh and low, and both returned 200.

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR