fix: strip reasoning parts when switching to non-interleaved models

[okossa]

Summary

Fixes cross-model switching errors when moving from a thinking model (e.g., Claude Opus) to a non-thinking model (e.g., GPT 5.2, Claude Sonnet).

• Strip reasoning parts from message history at the start of normalizeMessages() for models with interleaved: false
• Prevents API errors like messages.1.content.0.thinking: each thinking block must contain thinking

Fixes #11571

What Changed

packages/opencode/src/provider/transform.ts:

• Moved reasoning stripping logic to the beginning of normalizeMessages() so it runs before any model-specific early returns (Claude, Mistral, etc.)
• When all content is reasoning, inserts a placeholder text to avoid empty messages

packages/opencode/test/provider/transform.test.ts:

• Added 3 new tests for the cross-model reasoning stripping behavior
• Updated existing tests to reflect the new behavior

How I Verified It Works

1. Started OpenCode with Claude Opus (extended thinking enabled)
2. Got a response with thinking blocks
3. Switched to GPT 5.2 mid-session using Ctrl+M
4. Sent another message
5. ✅ No error - reasoning parts were stripped from history

Also ran the full test suite: bun test test/provider/transform.test.ts → 97 pass, 0 fail

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on the search results, here are the potentially related PRs (excluding the current PR #11572):

1. fix: strip incompatible thinking blocks when switching to Anthropic models #6748 - fix: strip incompatible thinking blocks when switching to Anthropic models

   • fix: strip incompatible thinking blocks when switching to Anthropic models #6748
   • Related: Addresses similar issue with thinking blocks when switching between models

2. fix: strip incompatible thinking blocks when switching to Claude #8958 - fix: strip incompatible thinking blocks when switching to Claude

   • fix: strip incompatible thinking blocks when switching to Claude #8958
   • Related: Directly related to stripping thinking blocks during model switching

3. fix(anthropic): ensure reasoning blocks precede tool_use in assistant messages #10474 - fix(anthropic): ensure reasoning blocks precede tool_use in assistant messages

   • fix(anthropic): ensure reasoning blocks precede tool_use in assistant messages #10474
   • Related: Addresses reasoning block handling in Anthropic models

These PRs appear to be related efforts to handle reasoning/thinking blocks across different models and contexts. PR #11572 seems to be a more comprehensive fix that handles stripping reasoning for all non-interleaved models, while the earlier PRs addressed specific provider cases.