Non-finite float metadata (inf/nan) is silently accepted and then dropped, losing metadata entries

[warter666]

What happened?

validate_metadata() accepts any Python float, including inf and nan. These values pass client-side validation, but they are not JSON-serializable, so they are silently dropped during persistence/serialization.

Consequences:

1. Silent metadata loss — adding {"score": float("inf"), "source": "a"} persists source but silently drops score; adding {"score": float("inf")} as the only key returns metadatas: [None] for that record. The data is gone after a client restart too.
2. Inconsistent validation surface — $gt/$lt filters against the dropped key silently miss the record, which is confusing to debug because the write appeared to succeed.

Reproduction (chromadb 1.5.9, PersistentClient)

import chromadb, tempfile

pc = chromadb.PersistentClient(path=tempfile.mkdtemp())
c = pc.get_or_create_collection("t")
c.add(ids=["1"], embeddings=[[0.1, 0.2]],
      metadatas=[{"score": float("inf"), "source": "doc-a"}])
print(c.get(ids=["1"])["metadatas"])
# [{'source': 'doc-a'}]   <- 'score' silently gone

Same for nan, for the update() path (validate_update_metadata), and for floats inside list-valued metadata (_validate_metadata_list_value).

Expected behavior

Writes containing non-finite floats should fail loudly with a ValueError at validation time, mirroring how nested dicts are already rejected. (Alternatively: document + sanitize, but rejecting seems consistent with the existing strictness.)

Happy to open a PR with the fix + tests if the approach sounds good.