Implement GPBFT message compression using zstd

[masih]

Add the ability to compress GPBFT messages controllable via manifest. Implement benchmarks to compare vanilla CBOR and ZSTD encoding.

Basic local run:

BenchmarkCborEncoding-12    	   47173	     25491 ns/op	  135409 B/op	      87 allocs/op
BenchmarkCborDecoding-12    	   64550	     18078 ns/op	   61728 B/op	     209 allocs/op
BenchmarkZstdEncoding-12    	   29061	     41489 ns/op	  193455 B/op	      88 allocs/op
BenchmarkZstdDecoding-12    	   66172	     17924 ns/op	  176517 B/op	     211 allocs/op

Fixes #786

[codecov]

Codecov Report

Attention: Patch coverage is 56.25000% with 28 lines in your changes missing coverage. Please review.

Project coverage is 69.31%. Comparing base (1b17c6c) to head (1578660).
Report is 5 commits behind head on main.

Files with missing lines	Patch %	Lines
msg_encoding.go	53.84%	12 Missing and 6 partials ⚠️
host.go	55.00%	5 Missing and 4 partials ⚠️
manifest/manifest.go	80.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #793      +/-   ##
==========================================
- Coverage   69.64%   69.31%   -0.33%     
==========================================
  Files          77       78       +1     
  Lines        7900     7953      +53     
==========================================
+ Hits         5502     5513      +11     
- Misses       1949     1979      +30     
- Partials      449      461      +12     

Files with missing lines	Coverage Δ
manifest/manifest.go	53.75% <80.00%> (-0.45%)	⬇️
host.go	65.15% <55.00%> (-2.29%)	⬇️
msg_encoding.go	53.84% <53.84%> (ø)

... and 3 files with indirect coverage changes

[Stebalien]

Lol, those decode benchmarks. Zstd is fast.

[masih]

We can reduce allocation significantly by pooling the buffers too across both CBOR and ZSTD. but baby steps/small PRs.

[rvagg]

out of interest - have you tried variations of the dst argument here? nil or oversizing by a % given your knowledge of the likely compression ratio?

[masih]

I have. My plan is to introduce pooling of buffers in a separate PR as it is an optimisation. There are a few places in encoding pipeline that we can benefit from it to reduce allocations.

In terms of impact, this probably doesn't matter as much considering the heap allocation numbers from passive testing. It is just wasteful.

[rvagg]

neat