Benchmark reports including comparisons from CI

[Ma27]

Description

Changes like #94 (and the previous ones) have shown that it's still a thing to underestimate the performance impact of a patch.

We should thus generate benchmark reports that are posted to each PR (including comparison to e.g. the previous commit or master).

[oberblastmeister]

I feel like the benchmarks could take way too long and I am not sure we can trust github actions to provide a consistent benchmarking environment. I think it would be better to run benchmarks on your own machine.

[Ma27]

Well, t his isn't supposed to be a 100% perfect environment for that, but I still think it's reasonable to to get some automated feedback if the situation is generally improving or getting worse. Also, an automated check won't decide whether something is mergeable or not, a human should always investigate anyways ;-)

[fogti]

[zseri@orca:~/devel/rnix-parser]$ git lfs checkout
Skipped checkout for "benches/all-packages.nix", content not local. Use fetch to download.                                                                                                                                                     
Checking out LFS objects: 100% (1/1), 803 KB | 0 B/s, done.

[nix-shell:~/devel/rnix-parser/src]$ cargo bench
[...]
all-packages            time:   [9.3041 us 9.4207 us 9.5911 us]
                        thrpt:  [13.026 MiB/s 13.261 MiB/s 13.428 MiB/s]
[...]
[nix-shell:~/devel/rnix-parser/src]$ git switch master
Previous HEAD position was ec908d6 Merge pull request #91 from darichey/rust-2021
Switched to branch 'master'
Your branch is up to date with 'origin/master'.
branchless: processing checkout

[nix-shell:~/devel/rnix-parser/src]$ cargo bench
[...]
all-packages/all-packages
                        time:   [6.9461 us 6.9626 us 6.9775 us]
                        thrpt:  [17.905 MiB/s 17.943 MiB/s 17.986 MiB/s]
[...]