feat(core) Calculate GPU timings using luma.gl QuerySet API

[ibgreen-openai]

GPUTimer Flow

1. Screen Pass opts.pass === 'screen'	➡️	2. Start Timestamp writeTimestamp(querySet, 0)	➡️	3. Render renderLayers(opts)	➡️	4. End Timestamp writeTimestamp(querySet, 1)
⬇️
Async Readback readTimestampDuration(0, 1)			➡️	Deck Metrics gpuTime + gpuTimePerFrame

Guardrails

Guardrail	Behavior
Screen pass only	Picking/offscreen passes are not timed
Timestamp support required	No-op when the device lacks timestamp-query
writeTimestamp required	No-op when the command encoder does not expose timestamp writes
One readback in flight	Avoids reusing the query set before results are consumed
Best-effort telemetry	Query failures are ignored instead of breaking render

Lifecycle

Event	Cleanup
Deck.finalize()	Destroys the GPU timer QuerySet and clears pending readback state

[coveralls]

coverage: 83.759% (+0.01%) from 83.746% — ib/gpu-timer into master