Pydantic converter with ToJsonOptions can't read back aliased models (pydantic >= 2.11) #2049

With ToJsonOptions set, PydanticJSONPlainPayloadConverter.to_payload serializes with
SchemaSerializer.to_json(value, exclude_unset=...), and from_payload validates with TypeAdapter.validate_json, which reads aliases. Since pydantic-core 2.33 (pydantic 2.11), SchemaSerializer.to_json no longer defaults to by_alias=True: it follows the model's serialize_by_alias, which is False by default. So on this path aliased fields are written under their Python names and then can't be read back. Without ToJsonOptions the converter uses pydantic_core.to_json, which still writes aliases, so only the options path is affected.

What happens:

• A model with Field(alias=...) or alias_generator=to_camel fails to decode. As an activity argument that is
  "Failed decoding arguments", a retryable ApplicationError, so it retries until schedule-to-close.
• A model with extra="allow" decodes, but the aliased field is silently lost (for example the OpenAI SDK's
  ResponseFunctionToolCall.async_ becomes None).

Who reaches it: anyone who sets ToJsonOptions, and every user of the Deep Agents, Google ADK and OpenAI Agents plugins in temporalio/ai-integrations, which each install a converter with ToJsonOptions(exclude_unset=True).

Repro (server-free): a model Order(order_id: str = Field(alias="orderId")) round-tripped through
PydanticPayloadConverter with ToJsonOptions(exclude_unset=True) raises ValidationError on pydantic 2.11+; with pydantic 2.10.6 it round-trips. pydantic_core.to_json writes orderId on all versions.

Proposed fix: pass by_alias=True on the options path, which matches to_json. I have a PR with a test ready.