gpu: snapshot per-batch LOD timing across batch boundary

[nclack]

Closes #154.

The per-level (LOD) timing in the metrics could read slightly wrong values. A batch's timing was read after it finished, but by then the next batch could have reused and overwritten those timers. These numbers are only for reporting, so written data was never affected — only the timing readout could be off.

Fix: give each in-flight batch its own set of timers (enough for the most that can be active at once), so a batch's timing is read from its own set rather than a later batch's. The ordering signal that had been sharing one of those timers is split out, leaving the timers purely for measurement. Batches that do no per-level work (a trailing partial write) are marked so the reader skips their timers instead of reading stale ones.

Reporting only; no change to written data.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/defs.limits.h	100.00% <ø> (ø)

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.