switch compression in zip archives to deflate

[syphar]

Up to discussion of course.

I realized:

• creating the archives can be quite slow, and the main issue is bzip compression speed
• Right now the webserver creates quite some CPU load, I guess partially because of bzip2 decompression.

Storage size is either nearly the same, at most deflate archives is 5-10% bigger.

After fixing the decompression benchmark we also were able to see the actual throughputs.

Then looking into benches:

regex html/compress bzip2
                        time:   [1.6310 s 1.6630 s 1.7063 s]
                        thrpt:  [6.0118 MiB/s 6.1685 MiB/s 6.2893 MiB/s]
                 change:
                        time:   [+0.4105% +2.4544% +5.4159%] (p = 0.06 > 0.05)
                        thrpt:  [−5.1376% −2.3956% −0.4088%]
regex html/decompress bzip2
                        time:   [103.30 ms 105.30 ms 107.35 ms]
                        thrpt:  [95.560 MiB/s 97.417 MiB/s 99.305 MiB/s]
                 change:
                        time:   [+828204% +841346% +854357%] (p = 0.00 < 0.05)
                        thrpt:  [−99.988% −99.988% −99.988%]
                        Performance has regressed.

regex html/compress deflate
                        time:   [66.063 ms 66.796 ms 67.226 ms]
                        thrpt:  [152.59 MiB/s 153.57 MiB/s 155.28 MiB/s]
                 change:
                        time:   [−2.8354% −1.1810% +0.2685%] (p = 0.19 > 0.05)
                        thrpt:  [−0.2677% +1.1951% +2.9181%]
regex html/decompress deflate
                        time:   [11.182 ms 11.482 ms 11.840 ms]
                        thrpt:  [866.41 MiB/s 893.38 MiB/s 917.35 MiB/s]
                 change:
                        time:   [+73719% +75869% +78203%] (p = 0.00 < 0.05)
                        thrpt:  [−99.872% −99.868% −99.865%]

I probably should have some more thinking / research / measurement on compression/decompression speed, and what the impact onto our webserver could be.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.