fix(core): add ChatBedrockConverse to deserialization allowlist

[Balaji Seshadri (gitbalaji)]

Summary

Fixes #34645.

• ChatBedrockConverse was absent from SERIALIZABLE_MAPPING in langchain_core/load/mapping.py.
• Since langchain-core>=1.2.5 (PR fix(core): serialization patch #34455 tightened the allowlist enforcement), calling langsmith.pull_prompt(..., include_model=True) with a ChatBedrockConverse model raises:

  ValueError: Trying to load an object that doesn't implement serialization:
  allowed_objects='core' only permits core langchain-core classes.
  

• ChatBedrock and BedrockLLM were already in the mapping — ChatBedrockConverse was simply missed.
• Fix: one entry added pointing ("langchain_aws", "chat_models", "ChatBedrockConverse") → langchain_aws.chat_models.bedrock_converse.ChatBedrockConverse (confirmed via ChatBedrockConverse.get_lc_namespace() returning ["langchain_aws", "chat_models"]).

Areas requiring careful review

• The serialization key ("langchain_aws", "chat_models", "ChatBedrockConverse") is derived from get_lc_namespace() + class name — confirmed against the source in langchain-ai/langchain-aws.
• No behaviour change for any other class.

Test plan

• test_chat_bedrock_converse_in_serializable_mapping — asserts the key is present and maps to the correct module path (no langchain-aws install required)
• Full tests/unit_tests/load/test_serializable.py suite passes (45 tests)
• ruff check, ruff format, mypy clean

AI disclaimer: This PR was developed with assistance from Claude Code (Anthropic).

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will degrade performance by 17.65%

⚠️ 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.

❌ 12 regressed benchmarks
✅ 1 untouched benchmark
⏩ 23 skipped benchmarks¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	test_import_time[HumanMessage]	239.3 ms	281.6 ms	-15.02%
❌	WallTime	test_import_time[ChatPromptTemplate]	549.8 ms	639.4 ms	-14.02%
❌	WallTime	test_import_time[tool]	473.5 ms	575 ms	-17.65%
❌	WallTime	test_import_time[InMemoryVectorStore]	552.5 ms	617.3 ms	-10.5%
❌	WallTime	test_import_time[LangChainTracer]	410.6 ms	474.2 ms	-13.41%
❌	WallTime	test_import_time[CallbackManager]	286.4 ms	331.3 ms	-13.55%
❌	WallTime	test_import_time[RunnableLambda]	442 ms	514.9 ms	-14.15%
❌	WallTime	test_import_time[Runnable]	444.5 ms	494.5 ms	-10.1%
❌	WallTime	test_import_time[PydanticOutputParser]	477.6 ms	553.7 ms	-13.75%
❌	WallTime	test_import_time[Document]	172.1 ms	197.1 ms	-12.67%
❌	WallTime	test_import_time[InMemoryRateLimiter]	158.6 ms	187.7 ms	-15.46%
❌	WallTime	test_import_time[BaseChatModel]	482.7 ms	573.5 ms	-15.82%

---

_{Comparing gitbalaji:fix/core-chat-bedrock-converse-allowlist (0f1fc20) with master (2d1492a)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 (0b975d4) during the generation of this report, so 2d1492a was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[Balaji Seshadri (gitbalaji)]

CodSpeed regressions are false positives

All 7 flagged benchmarks are unrelated to this change.

Our change: a single static dict entry added to SERIALIZABLE_MAPPING in langchain_core/load/mapping.py. No logic, no imports, no runtime code paths affected.

Why none of these benchmarks can regress from our change:

• test_import_time[RunnableLambda/CallbackManager/Runnable/tool/PydanticOutputParser/ChatPromptTemplate] — these measure subprocess import times for completely unrelated modules. mapping.py is not imported as part of any of these import chains (it's only loaded on demand during deserialization). A static dict entry cannot affect import time of RunnableLambda.
• test_async_callbacks_in_sync — benchmarks async streaming callbacks. No relationship to deserialization mapping.

The consistent ~10–15% regression across 6 unrelated import benchmarks in a single run is a textbook signature of a noisy shared runner, not a real regression. CodSpeed itself warns: "Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data."

[ccurme (ccurme)]

Duplicated with #34510.