fix(chroma): cap hnsw:num_threads to available CPU cores during collection load Closes langchain-ai/langchain#32678

[AshutoshRavindraIwale]

Summary

This PR adds a guard in the Chroma vectorstore implementation to ensure that the hnsw:num_threads metadata parameter does not exceed the number of available CPU cores when loading a persisted collection.

Context

Fixes a bug where moving a Chroma collection from a server with N CPU cores to one with fewer cores (M < N) could cause a ValueError if hnsw:num_threads was set higher than supported on the new host. See: langchain-ai/langchain#32678

Changes

• In langchain_community/vectorstores/chroma.py, thread allocation in collection metadata is capped to os.cpu_count() if necessary during init.
• Adds/updates tests in tests/unit_tests/vectorstores/test_chroma.py:
  • Verifies capping of excessive thread settings.
  • Ensures valid/low thread counts remain untouched.
  • Checks handling for missing/None collection metadata.
  • Confirms backward compatibility with ChromaDB integrations.

Testing

• All tests, including new edge cases, pass locally.
• Real and mock ChromaDB tested for backwards compatibility.

---

Closes langchain-ai/langchain#32678