fix: use uuid4 (not uuid1) for auto-generated ids in async constructors

[Andrew Chen (chuenchen309)]

Problem

PGVector's sync constructor path (__from, used by from_texts /
from_embeddings / from_documents) generates auto-assigned document ids with
uuid.uuid4():

# __from
if ids is None:
    ids = [str(uuid.uuid4()) for _ in texts]

The async path (__afrom, used by afrom_texts / afrom_embeddings /
afrom_documents) instead used uuid.uuid1():

# __afrom
if ids is None:
    ids = [str(uuid.uuid1()) for _ in texts]

uuid1 encodes the host's MAC address and a timestamp. So when a caller
lets the store auto-generate ids (the common case), the async API:

1. produces ids that are inconsistent with the random ids the sync API
   produces for the same call, and
2. leaks the server's hardware address and record-creation time into the
   ids that get stored in the DB and returned to callers.

These ids are not discarded — __afrom passes them into aadd_embeddings,
which writes them as the primary-key id. aadd_embeddings itself already uses
uuid4 for its own auto-generation, so this lone uuid1 looks like an
oversight rather than a deliberate choice.

Fix

Use uuid.uuid4() in __afrom, matching __from and aadd_embeddings.

Test

Added tests/unit_tests/v1/test_afrom_uuid.py, a pure-Python test (no live
Postgres): constructing with async_mode=True defers __post_init__ so no
connection is opened, and aadd_embeddings is stubbed to capture the generated
ids. It asserts every auto-generated id is a UUIDv4. Fails before the change
(version == 1), passes after.

$ python -m pytest tests/unit_tests/v1/test_afrom_uuid.py -q
1 passed

ruff check passes on the changed files.

---

Disclosure: I used AI assistance (Claude) to help locate this sync/async
parity gap and draft the test. I reviewed the change, ran the test suite, and
take responsibility for its correctness.