Fix JsonlWriter crash on pandas.Timestamp metadata

[discobot]

Fixes #325.

orjson serializes plain datetime/date/time natively but deliberately rejects their subclasses, and pandas.Timestamp (what ParquetReader yields for timestamp columns) is a datetime.datetime subclass — with no default= handler, orjson.dumps raises. The handler added here emits isoformat() for date/time subclasses, matching orjson's native datetime output exactly, while other unserializable types still fail loudly (vs. a blanket default=str, which would silently stringify anything — happy to switch to that if preferred).

The new regression test fails on main with the exact TypeError: Type is not JSON serializable: Timestamp from the issue and passes with this change.

---

Note

Low Risk
Narrow change to JSON serialization in JsonlWriter with explicit errors for unknown types; no auth, security, or data-path changes beyond fixing a metadata edge case.

Overview
Fixes JsonlWriter crashing when document metadata contains pandas.Timestamp (and other date/time subclasses) by passing a default=_json_default hook to orjson.dumps. Subclasses that orjson won’t serialize natively are written as isoformat() strings, aligned with native datetime output; other non-JSON types still raise TypeError.

Adds a regression test that writes a document with timestamp metadata and asserts the ISO string in the JSONL output, plus a require_pandas skip decorator for optional test deps.

^{Reviewed by Cursor Bugbot for commit a555fec. Bugbot is set up for automated code reviews on this repo. Configure here.}