Fix inaccurate measurements after tracker restarts

On startup `flux-commit-tracker` can measure an already-applied revision using its latest reconciliation timestamp instead of its original apply time. This causes a commit that got applied in a typical, expected amount of time to be reported again with a highly inflated cycle time due to it using the current, rather than original, apply time of that OCI revision. This distorts the accuracy of the metrics and can even trigger false-alarm alerts. Also, retries etc can end up reading a newer revision or timestamp than the update that originally queued the measurement.

To address these issues we should:

- Skip measurements for existing revisions discovered at process startup
- Preserve the revision and original apply timestamp throughout processing and retries