Switch compression algorithm from zstd to lz4

[emilk]

Advantages:

• Faster compilation times
• 3x faster encode and decode
• Pure Rust crate
• Can compress in wasm

Disadvantages:

example .rrd	zstd	lz4
api_demo	48 kB	93 kB
car	120 kB	375 kB
clock	28 kB	53 kB
colmap	227 MB	241 MB
deep_sdf	19 MB	20 MB
dicom	39 MB	64 MB
nyud	535 MB	634 MB
plots	89 kB	163 kB
raw_mesh	1.5 MB	5.6 MB
text_logging	1.9 kB	3.0 kB

I'm not sure what is the best trade-off here. I'm gonna check the compilation times too.

Note that these compilation times impact our Rust users, as well as our contributors.

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct

PR Build Summary: https://build.rerun.io/pr/2112

[emilk]

Compilation times

TL;DR: nothing won 😭

sccache --stop-server && rm -rf /Users/emilk/Library/Caches/Mozilla.sccache && sccache --start-server
cargo clean && cargo build --timings -p rerun

Before (main)

478 deps in 2m 12s

After

(together with EmbarkStudios/puffin#135)
470 deps in 2m 12s

[Wumpf]

It's strange that re_log_types took so much longer on the lz4 run. It also seems like the whole thing is hitting some bottleneck where it can't parallize - it would be that this is still a win on a machine with less cores than yours, i.e. a strong improvement for many users. Might be worth doing a comparison run that is limited to 4 cores 🤔

[emilk]

We've decided to prioritize CPU use right now over disk space. We may revisit this in the future, allowing users to pick different compression algorithms and levels.