fix: honor top-level reasoning `none` for supported Groq models (#19361)

## Background

Groq omitted top-level `reasoning: 'none'`, causing supported Qwen
models to retain their default reasoning behavior and consume output
tokens instead of directly returning answer text.

## Root Cause

The Groq adapter explicitly excluded `reasoning: 'none'` from unified
reasoning mapping. The reproduction confirmed that omission left
reasoning enabled, while explicitly sending provider-specific
`reasoningEffort: 'none'` disabled it.

## Summary

Added model-aware mapping of top-level `reasoning: 'none'` to
`reasoning_effort: 'none'` for `qwen/qwen3.6-27b`, warnings for
unsupported models, and preserved provider-option precedence. Updated
the model union, provider documentation, live example, regression tests,
and patch changeset while removing stale runtime handling for deprecated
`qwen/qwen3-32b`.

## Testing

Regression coverage verifies Qwen 3.6 disablement, unsupported GPT-OSS
warnings, and provider-option precedence with distinct values. Node and
Edge tests each passed all 43 focused tests, and package/workspace type
and repository checks passed.

## End-to-end Validation

- The recorded reproduction command exited zero without the reproduction
signal after sending `reasoning_effort: 'none'`.
- `pnpm -C examples/ai-functions exec tsx
src/stream-text/groq/reasoning-effort.ts` returned answer text without
entering the reasoning stream.

## Related Issues

Fixes #19357

Closes #19359

---------

Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: gr2m <39992+gr2m@users.noreply.github.com>
Co-authored-by: Nick Oates <58091943+n1ckoates@users.noreply.github.com>