fix(core): prevent cross-type content block merging in merge_lists

[Mason Daugherty (mdrxy)]

merge_lists merges dicts by matching index values but does not check type. When a model returns both text and tool calls in a streaming response with output_version="v1", text blocks (index: 0 from the streaming wrapper) and tool_call_chunk blocks (index: 0 from the API) merge into a single corrupt dict with type: "text" carrying orphaned name/args/id fields. The init_tool_calls validator cannot find type: "tool_call_chunk" to promote, so tool calls silently vanish from content_blocks.

This affects BaseChatOpenAI subclassers without registered block translators and any model that returns text alongside tool calls (e.g., GLM-5 via Baseten). Related to #32888 which proposes the broader fix of raising on type conflicts in merge_dicts.

Changes

• Add a type guard to merge_lists index-matching: prevent merging when one block has type: "text" and the other does not — a text-vs-everything-else partition that keeps text and tool_call_chunk blocks separate so init_tool_calls can promote them correctly. Non-text blocks at the same index (e.g., Anthropic's tool_use + input_json_delta) still merge as before.

---

Revert basetenlabs/langchain-baseten#6 when released

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 13 untouched benchmarks
⏩ 23 skipped benchmarks¹

---

_{Comparing mdrxy/fix-merge-list (d8537d1) with master (69a7b9c)²}

Footnotes

1. 23 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on master (2bad58a) during the generation of this report, so 69a7b9c was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩