[Metrics] Add preemption metrics and accumulation logic

[PauloVLB]

b/538558303

Note: this PR is chained with #5394

Problem

When a preemption occurs, we were only recording the duration of the active fuzzing round as wasted time. If a task is preempted mid-way, all previous successful rounds in that same task are also completely wasted since the task never reaches postprocess. This leads to a significant underestimation of wasted time.

Proposed Solution

Enhanced time tracking to capture the waste across the entire fuzzing session while maintaining metric accuracy.

• Added FUZZER_PREEMPTED_TOTAL_FUZZ_TIME and JOB_PREEMPTED_TOTAL_FUZZ_TIME metrics to monitoring_metrics.py.
• Enhanced _TrackFuzzTime in fuzz_task.py to track accumulated wasted time across all rounds of a session (_accumulated_wasted_time).
• Upon preemption, the callback now emits the Total Accumulated Session Time (Current Round + Previous Rounds) to preempted_total_time.
• To fix subtraction math in dashboards, the callback also emits the duration of the interrupted round only to total_time, ensuring it reflects the attempted time without gaps.
• Added tracker.preempted flag to skip metric emission in __exit__ for interrupted rounds and avoid double counting.

Validation

Validated in dev environment.

[dylanjew]

This doesn't include build set up and corpus sync time, which is also effectively wasted time from a fuzzing perspective. But our total fuzz time metrics also don't include that, so I understand that we may want it to match.

I don't know how long that set up takes, but it might be worth tracking that as well.

[dylanjew]

Also, not blocking at all because any metrics here will help but definitely worth discussing/thinking about

[PauloVLB]

Yes, it was done that way to match the existing metric, so the subtraction (total_time - preempted_time) makes sense.

Measuring the duration of the full main stage (including build set up and corpus sync) is the next step I'm working on. Thanks for pointing that out!