Move Cleanup Backfills after main SynchronizerCycle & add workers pool

[akremsa]

What this PR does / Why we need it:
During the contributor's meeting, we agreed with @Laremere to follow this pattern in the cleanup backfills process.
Move CleanupBackfills after main SynchronizerCycle.

Special notes for your reviewer:
There are some benchmarks:

Using workers pool
workers=5; expired backfills=50
BenchmarkCleanupBackfills-12 40 28685354 ns/op 15493765 B/op 90618 allocs/op

workers=3; expired backfills=50
BenchmarkCleanupBackfills-12 34 29910474 ns/op 15493394 B/op 90614 allocs/op

workers=1; expired backfills=50
BenchmarkCleanupBackfills-12 26 40829001 ns/op 15492384 B/op 90604 allocs/op

workers=5; expired backfills=200
BenchmarkCleanupBackfills-12 9 114774808 ns/op 61991684 B/op 362357 allocs/op

workers=3; expired backfills=200
BenchmarkCleanupBackfills-12 8 130279702 ns/op 61985000 B/op 362315 allocs/op

workers=1; expired backfills=200
BenchmarkCleanupBackfills-12 7 200287135 ns/op 61978501 B/op 362230 allocs/op

Previous straightforward implementation
expired backfills=50
BenchmarkCleanupBackfills-12 25 41920697 ns/op 15491571 B/op 90599 allocs/op

expired backfills=200
BenchmarkCleanupBackfills-12 6 186510753 ns/op 61971938 B/op 362200 allocs/op

As you can see, results with worker=1 are pretty close to not using the worker at all.
Also, there is no big difference between 5 and 3 workers, taking into account that all benchmarks have a margin of error.

Closes #1324 .

[akremsa]

@Laremere, @yeukovichd let's discuss pros and cons of putting cleanup method at the beginning or at the end of a sync cycle.
We agreed to put it at the beginning because this approach guarantees that we work only with relevant backfills and non of them will be deleted during MMF execution.
We had lots of issues with tests where cleanup, when it was put at the end, is involved. It's really hard to predict how long will it take to delete everything and introducing lots of sleep calls brings more complexity and doesn't make this whole sync and testing process stable.

[aLekSer]

Also this is discussed here #1324 in more details.

[Laremere]

The reason to move it to after the cycle is because that's when the synchronizer isn't being waited upon, and is doing little work. The backfill will have been a part of many cycles before the cleanup, so I don't see improving by one cycle to be that helpful.

The testing problem can be solved by:

1. Create backfill
2. Wait till it should expire
3. Call fetch matches, having mmf and evaluator do nothing.
4. Call fetch matches again
5. backfill should be gone.

Since backfill cleanup will finish before the next fetch matches can run, running fetch matches a second time guarantees that the cleanup which should clean up the backfill has finished.

[aLekSer]

/gcbrun

[aLekSer]

/gcbrun

[akremsa]

@Laremere please review this PR again

[aLekSer]

Now CleanupBackfills is performed at then end of the regular Synchronizer cycle. Regular Synchronizer steps do not depend on Cleanup running time, possible bugs in Cleanup, etc.

[aLekSer]

/gcbrun