TestHasIsBloomCached failures

[Stebalien]

Moved from: ipfs/kubo#3208

--- FAIL: TestHasIsBloomCached (0.03s)
    bloom_cache_test.go:100: Bloom filter has cache miss rate of more than 5%
FAIL
FAIL    github.com/ipfs/go-ipfs/blocks/blockstore   0.053s

[taylormike]

I believe I found the root cause for this test failure. It appears to be a race condition/ timing issue that sometimes causes 'BloomActive()'s' Goroutine to read b.active prior to b.active's assignment via 'Rebuild()'s' Goroutine.

I was able to consistently reproduce this test failure after adding a 300 Microsecond sleep in Rebuild.
(See code snippet below)

@Stebalien @Kubuxu @lgierth, @whyrusleeping Thoughts?

Rebuild's() Goroutine:

        close(b.rebuildChan)
	time.Sleep(300 * time.Microsecond) //<-- Temp Code that reproduces the test failure
	atomic.StoreInt32(&b.active, 1)

BloomActive's() Goroutine:

func (b *bloomcache) BloomActive() bool {
	return atomic.LoadInt32(&b.active) != 0
}

[whyrusleeping]

@taylormike does it happen if you put that sleep specifically there? Or does it also work if you put the sleep before the channel close?

[taylormike]

@whyrusleeping

1. The test fails if I put the sleep specifically there. (one sleep)
2. The test passes if I put the sleep before the channel close. (one sleep)
3. The test fails if I put the sleep there and I put the sleep before the channel close. (two sleeps)