[pull] main from temporalio:main - #189

* fix(contrib/pydantic): reuse TypeAdapters across payloads

PydanticJSONPlainPayloadConverter.from_payload constructed a fresh
pydantic TypeAdapter for every payload, rebuilding the core schema each
time for non-class hints such as discriminated unions and generic
collections.

Cache adapters per converter instance, keyed on hashable type hints;
unhashable hints keep constructing fresh adapters. The cache is
unbounded by default and configurable via the new keyword-only
max_cached_type_adapters option on PydanticJSONPlainPayloadConverter
and PydanticPayloadConverter (positive bounds with LRU eviction, zero
disables caching, negative raises ValueError).

Fixes #1695

* fix(contrib/pydantic): default type adapter cache bound to 1024

Bound the per-converter type adapter cache to 1024 entries by default
with LRU eviction, capping worst-case memory even with runtime-generated
hints while never evicting for typical static hint sets. None remains
available for an unbounded cache and zero still disables caching.

* test(contrib/pydantic): cover re-imported class cache isolation

The workflow sandbox re-imports user modules, producing distinct class
objects with identical names. Verify each gets its own cache slot and
validates to its own world's class, even when one converter is shared.

* fix(contrib/pydantic): avoid double hash on cached decode path

Address review: try the cache directly instead of pre-hashing every
hint. On TypeError, hash the hint once only to distinguish an
unhashable hint (bypass the cache with a fresh adapter) from a
TypeError raised during adapter construction (re-raise), keeping
adapter errors unsuppressed.