Fix flaky test TestChaosUpload

[Joerger]

Change:

• Make the test run magnitudes faster and allow the --race flag.
  • From my local testing, it doesn't not seem like the slow scan period was not warranted and may have just been covering up underlying issues with the test. That may change after seeing CI run it.
• Don't treat empty uploads as a sessionError in uploadEncrypted, mirroring the default upload method.
  • It's unclear whether this should actually be working this way in either case. We have had issues in the recent past where recordings are truncated and never cleaned up - e.g. moved to the corrupted folder with an .error file if relevant. However, this can be handled in a follow up PR if the legacy behavior is deemed undesirable rather than blocking this high priority flaky test fix.
• Add a missing FS unlock when uploadEncrypted hits an unexpected error. This flake appeared once ^ was fixed.

Closes #33099

[Joerger]

@rosstimothy @zmb3 Can you give this a flaky test skip? Looks like it's just timing out, takes just under 6s locally:

> go test -run ^TestChaosUpload$ github.com/gravitational/teleport/lib/events/filesessions -tags=libfido2,piv --count=1 --race              
ok  	github.com/gravitational/teleport/lib/events/filesessions	5.515s

[rosstimothy]

/excludeflake *

[backport-bot-workflows]

@Joerger See the table below for backport results.

Branch	Result
branch/v18	Failed

[ravicious]

Is it feasible to backport these changes to v17? I just got a hit on v17. I tried backporting this PR myself, but it doesn't backport cleanly.