Cache compression

[mrowqa]

Based on #223. Some initial implementation of compression for cache system.

Things to discuss:

• configuring compression level - I have temporarily added command line option for it, but it looks like too many details. It'd be probably better to have some performance presets for wasmtime (mentioned in Relocation of memory.grow and memory.size #199),
• compression algorithm, currently using zstd
  • I preferred using official implementation with Rust bindings rather than some custom implementation; currently we require cmake and clang anyway, but we will probably want to not depend on them in future,
  • I did a quick research on different algorithms and other good choice could be lzma2. Anyway, it's hard to compare usage of specific algorithms without good benchmarks on our specific use case. I just do some quick test runs for having an approximated comparison.

[mrowqa]

Some perf runs:

• no compression (basically the Add command line option for custom cache directory #223)
• zstd, lvl=3 (default)
• zstd, lvl=5
• zstd, lvl=10
• zstd, lvl=20

With such comparisons it'd be nicer to have some chart. Anyway, looking at bigger files:

• start without cache: lvl=3;5 not noticeable, lvl=10 very little overhead, lvl=20 big overhead
• start with cache: lvl=3;5 visible overhead, but still not that big, lvl=10 visible improvement to previous one, still slower than no compression, lvl=20 again, visible improvement to previous level, still slower than without cache (note: the cache were probably read from OS filesystem cache; "hot buffer" in terminology of databases)
• compression ratio: more or less 10% +-5%, each tested level introduced noticeable improvements. Cache without compression has low-mid entropy (what felt strange at first glance since usually code has high entropy).

[mrowqa]

Rebased onto #230. I haven't removed the config option for compression level, but we'll do it soon anyway, when working on the cache daemon.

[sunfishcode]

Looks good!