fix: forward the same model.parameters keys as the Python SDK

[apucacao]

The JS and Python SDKs have to forward the same model.parameters keys before 1.0. Jeff's review of launchdarkly/python-ai-sdk#107 found that the lists from #73 differ from Python's. The shared lists are now in the cross-SDK spec (launchdarkly/ai-sdks-monorepo#42, TESTING.md §1.12), and Python is moving to them too. This PR makes each JS handler forward exactly those lists.

What changed:

• Claude Messages: stops forwarding container, inference_geo and metadata.
• OpenAI Messages: stops forwarding context_management, include, instructions, metadata, moderation, prompt_cache_retention, safety_identifier, store, truncation and user, on both invoke and stream. max_tool_calls is not forwarded: the openai package has no such field on ResponseCreateParamsBase, so the spec cuts it for both SDKs.
• OpenAI Agents (handler and native graph): stops forwarding context_management, prompt_cache_retention, store and truncation.
• LangChain ChatOpenAI (both LangChain packages and the LangGraph native graph's default model): stops forwarding audio, disable_streaming, metadata, modalities, output_version, prompt_cache_key, prompt_cache_retention, reasoning, stream_usage, streaming, tags, use_responses_api, user, verbose and zdr_enabled. prompt_cache_key is cut because the Python ChatOpenAI has no such field; OpenAI Messages still forwards it.
• LangChain ChatAnthropic: stops forwarding context_management, disable_streaming, inference_geo, metadata, output_version, stream_usage, streaming, tags and verbose. It now takes stop as stopSequences and max_tokens_to_sample as maxTokens, because ChatAnthropic has neither field under those names.
• LangChain ChatBedrockConverse: stops forwarding disable_streaming, guardrail_config, metadata, output_version, stream_usage, streaming, supports_tool_choice_values, tags and verbose.
• Claude Agents and Vercel: no change.

Following the spec, nested values with the wrong shape are dropped instead of sent. Examples are a thinking without a type, a reasoning that is not an object, or a tool_choice of the wrong type. OpenAI Messages reasoning keeps only effort and summary, as OpenAI Agents already did.

Testing:

• Each handler has a test that the forwarded keys are exactly the spec's list, and that invoke and stream forward the same keys and values.
• Every removed key is in that package's __tests__/never-forwarded.ts, so the existing bag tests prove it is gone.
• New tests cover dropping malformed nested values.
• yarn test, yarn typecheck and biome check . pass locally.

---

Note

Overview
Aligns JS provider handlers with the cross-SDK model.parameters allowlists (TESTING.md §1.12) so forwarding matches the Python SDK before 1.0.

Handlers now forward only generation-shaping keys and stop passing retention, attribution, server-side state, and wiring fields that were previously allowed (e.g. Claude Messages: container, inference_geo, metadata; OpenAI Messages/Agents: store, context_management, truncation, etc.; LangChain: reasoning, use_responses_api, guardrail_config, metadata, streaming flags, and more). Malformed nested values (thinking, tool_choice, reasoning, output_config, etc.) are dropped instead of sent; OpenAI Messages trims reasoning to effort/summary. LangChain ChatAnthropic maps max_tokens_to_sample → maxTokens and stop → stopSequences. OpenAI Agents removes context_management transformation from buildModelSettings.

Tests add canonical-key parity for invoke vs stream, expand NEVER_FORWARDED_KEYS, and cover malformed-parameter drops.

^{Reviewed by Cursor Bugbot for commit 0e963d3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[apucacao]

bugbot run

[cursor]

Stale Bugbot comment from a previous run.

[apucacao]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 0e963d3. Configure here.}