Use hashbrown for better performance

[jmmv]

Profiling sandboxfs during large builds showed a non-negligible (but not large) amount of time going into hashtable operations. Given that we don't care about "secure" hashing, we can afford to switch to a faster implementation like hashbrown, so do it here.

This yields a ~1% performance boost during the Bazel build of a large application using sandboxfs. The following numbers are based on 10 consecutive builds for each case and are in seconds:

std: mean 3429.56, median 3425.00, stddev 21.56 hashbrown: mean 3394.50, median 3378.50, stddev 36.31

[djc]

(You're probably aware of this already, but in case anyone else is wondering about this, you actually switched from std::collections::HashMap with SipHash 1-3 to hashbrown::HashMap with FxHash -- so it's unclear whether the performance came from the change of implementation, the change of hash function, or (most likely) both.)

[jmmv]

@djc Ah thanks. I noticed that I could make the standard HashMap use FxHash, but the numbers in the hashbrown page seemed convincing to switch both details at once :P I could measure again all these cases, but given that the performance differences are pretty minor and there is some noise in the metrics, it'd be hard to appreciate those.

[djc]

Yeah, it makes sense to switch both at the same time in this case -- it's just that your commit message saying "Given that we don't care about "secure" hashing, we can afford to switch to a faster implementation like hashbrown, so do it here." does not entirely make sense.

[jmmv]

Thanks @hlopko. Also updated the commit based on @djc's comment.