store: Preallocate output buffer when encoding postings.

[pstibrany]

This PR modifies diffVarintSnappyEncode method to preallocate output buffer based on number of passed postings. This helps to avoid reallocations.

name                      old time/op    new time/op    delta
EncodePostings/10000-4       114µs ± 1%     106µs ± 1%   -6.55%  (p=0.009 n=3+9)
EncodePostings/100000-4     1.11ms ± 1%    1.05ms ± 0%   -5.39%  (p=0.007 n=3+10)
EncodePostings/1000000-4    11.6ms ± 2%    10.7ms ± 0%   -7.21%  (p=0.009 n=3+9)

name                      old alloc/op   new alloc/op   delta
EncodePostings/10000-4      48.5kB ± 0%    13.6kB ± 0%  -71.98%  (p=0.000 n=3+10)
EncodePostings/100000-4      506kB ± 0%     131kB ± 0%  -74.08%  (p=0.000 n=3+10)
EncodePostings/1000000-4    5.86MB ± 0%    1.25MB ± 0%  -78.62%  (p=0.000 n=3+10)

name                      old allocs/op  new allocs/op  delta
EncodePostings/10000-4        18.0 ± 0%       2.0 ± 0%  -88.89%  (p=0.000 n=3+10)
EncodePostings/100000-4       26.0 ± 0%       2.0 ± 0%  -92.31%  (p=0.000 n=3+10)
EncodePostings/1000000-4      36.0 ± 0%       2.0 ± 0%  -94.44%  (p=0.000 n=3+10)

Notes for reviewers: I chose to pass expected number of postings as another parameter. Alternative I considered was introducing an interface with PostingsCount method, which may look nicer, but would add hidden dependency into diffVarintSnappyEncode -- for postings implementing the interface, it would be faster (use less allocs) than for postings not implementing it. Passing extra parameter is more explicit.

• [NA] I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

Verification

Added benchmark.

[pracucci]

Results are amazing! Good job 👏 I left a minor nit, but no blocker to me.

[bwplotka]

💪

Motto for the week: PREALLOCATE ;p

[pstibrany]

💪

Motto for the week: PREALLOCATE ;p

Indeed :)

[pstibrany]

Test and doc failures seem to be unrelated to my changes :-(

[bwplotka]

Thanks!