test(openlineage): Make test_ordering deterministic instead of sleeping - #6791

TestGetRuns::test_ordering asserts the most recently upserted run sorts first. get_runs() orders by updated_at, which each upsert stamps as int(time.time() * 1000) — millisecond resolution, with no secondary sort key. The test slept time.sleep(0.01) between the two upserts so r2's timestamp would exceed r1's.

On a coarse-clock platform (e.g. ~15.6 ms timer granularity) a 10 ms sleep can fail to advance the millisecond, so both runs get the same updated_at, the tie falls back to insertion order, and r1 sorts first — the assertion flakes.

This replaces the sleep with a monotonic clock injected via monkeypatch, so each upsert gets a strictly increasing timestamp deterministically — no wall-clock dependence, no sleep.

Verified against the real OpenLineageStore SQL: with a same-millisecond clock the old test returns r1 (reproduces the flake); with the injected clock it returns r2 on 200/200 runs.

Release note: NONE

_{Developed with AI assistance (Claude Code); I directed, reviewed, and verified it locally.}