Use serde_bytes to reduce (de)serilization speed of 8-bit filters by 96%

[JanBerktold]

I noticed that deserialization of the BinaryFuse8 using serde takes a long time. In my use case with 450 millions members, it takes 2-3 minutes on a modern machine. The majority of this time is spent on the byte-by-byte deserialization of the fingerprints slice. The specialization offered by the serde_bytes crate solves that exact problem.

I also added a benchmark to show the difference and posted the results from a run on an M2 MacBook below.

Before:

BinaryFuse8/serialize/500000
                        time:   [263.42 µs 269.47 µs 272.86 µs]
BinaryFuse8/deserialize/500000
                        time:   [1.2453 ms 1.2498 ms 1.2553 ms]

After:

BinaryFuse8/serialize/500000
                        time:   [9.4238 µs 9.4404 µs 9.4479 µs]
                        change: [-96.524% -96.466% -96.401%] (p = 0.00 < 0.05)
                        Performance has improved.
BinaryFuse8/deserialize/500000
                        time:   [14.670 µs 14.727 µs 14.801 µs]
                        change: [-98.822% -98.816% -98.809%] (p = 0.00 < 0.05)
                        Performance has improved

[JanBerktold]

Hi @ayazhafiz, could you take a look at this one please?

[ayazhafiz]

Thank you! Sorry for the delay.

[ayazhafiz]

Do you mind pushing an empty commit? I don't see a way to retry the CI. I think it's a flake.

[JanBerktold]

Sure thing -- done.