Fix stale mapping and respect delete in rsync_from_disk

[PauloVLB]

b/530149063

Problem

ProtoFuzzTargetCorpus.rsync_from_disk keeps a stale internal mapping of files to delete. Since the corpus_pruning task calls this method twice (once after initial pruning and again after cross-pollination), the second pass attempts to delete the exact same files again, causing a flood of 404 warnings. It was also completely ignoring the delete argument.

Link for logs

Proposed Solution

Update the internal mapping after deletion to remove the files that were just deleted. Also, wrap the deletion logic to respect the delete flag.

Validation

Added a new unit test test_rsync_multiple_calls in corpus_manager_test.py to confirm that multiple calls don't try to delete the same files.

Also validated in dev environment: those are logs from a corpus_pruning task successfully executing without the mentioned warnings.

Link

Why weren't these warnings visible before?

The bug was always there but silent. Before switching to threads, we used processes. On Linux, the background thread responsible for flushing logs to Cloud Logging dies in forked child processes. Thus, warnings from child processes (like these 404s) were swallowed. Threads keep execution in the same process, making the warnings visible.