perf: coalesce empty checkpoints in batch

[yaythomas]

Issue #, if available:
closes #325

Description of changes:
Taking over this excellent PR from #336, just to fix some minor formatting issues. Didn't have push perms on contributor branch.

Empty checkpoints (used by map/parallel branch resubmitters when resuming from timed waits) no longer count toward the 250-operation batch limit beyond the first. This prevents 300+ concurrent branch resumes from splitting across multiple API batches.

from original PR:
Implements the empty-checkpoint coalescing optimization described in #325.

When many concurrent map/parallel branches resume from a wait operation simultaneously, each branch's resubmitter enqueues an empty checkpoint (no OperationUpdate). Previously, these counted toward the 250-operation batch limit, causing 300 concurrent branches to produce 2 API calls (250 + 50) instead of 1.

• Tracks effective_operation_count separately from len(batch)
• The first empty checkpoint in a batch counts as 1 effective operation
• All subsequent empty checkpoints in the same batch are coalesced in without incrementing the effective count
• Applied consistently across all three collection phases: overflow queue drain, first-op blocking wait, and time-window loop
• Updated debug logging to clearly show real ops sent vs empties excluded

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[yaythomas]

not necessary. got push perms on o.g pr