Adds configurable compression algorithms for chunks

[cyriltovena]

This introduces a new chunk format (v2) that contains the current compression algorithm used to encode the chunk.

Loki can still read (v1) format which are by default gzip, and the new format is now the default with still gzip as encoding.

You can mix multiple encodings over time now, if you want to try different format.

We support those encoding now:

• gzip: default gzip golang implementation the slowest with the best compression ratio.
• gzip-1: gzip with best speed configuration, a bit fast but also a lower compression ratio.
• lz4 : fast compression and decompression speed. (use more memory than snappy)
• none: no compressoion
• snappy: the popular google snappy compression format
• snappyv2: an optimized version of snappy see https://github.com/klauspost/compress/tree/master/s2

Chunk size for each

  --- PASS: TestChunkSize/none (0.01s)
        Chunk size 2.7 MB
       characters  2623262
    --- PASS: TestChunkSize/gzip (0.01s)
        Chunk size 144 kB
        characters  2623262
    --- PASS: TestChunkSize/gzip-1 (0.01s)
         Chunk size 158 kB
        characters  2623262
    --- PASS: TestChunkSize/lz4 (0.00s)
        Chunk size 188 kB
        characters  2623262
    --- PASS: TestChunkSize/snappy (0.00s)
        Chunk size 272 kB
       characters  2623262
    --- PASS: TestChunkSize/snappyv2 (0.00s)
        Chunk size 205 kB
        characters  2623262

Decompression speed benchmark

BenchmarkRead/none-16         	      15	  71535984 ns/op	134484800 B/op	  491005 allocs/op
--- BENCH: BenchmarkRead/none-16 bytes per second  1.8 GB
BenchmarkRead/gzip-16         	       5	 212703812 ns/op	134582374 B/op	  491465 allocs/op
--- BENCH: BenchmarkRead/gzip-16 bytes per second  617 MB
BenchmarkRead/gzip-1-16       	       5	 221152209 ns/op	134588478 B/op	  491452 allocs/op
--- BENCH: BenchmarkRead/gzip-1-16 bytes per second  593 MB
BenchmarkRead/lz4-16          	      10	 100376532 ns/op	141235551 B/op	  491049 allocs/op
--- BENCH: BenchmarkRead/lz4-16  bytes per second  1.3 GB
BenchmarkRead/snappy-16       	      10	 101250653 ns/op	134560112 B/op	  491038 allocs/op
--- BENCH: BenchmarkRead/snappy-16 bytes per second  1.3 GB
BenchmarkRead/snappyv2-16     	      10	 108484452 ns/op	137413780 B/op	  491057 allocs/op
--- BENCH: BenchmarkRead/snappyv2-16 bytes per second  1.2 GB
PASS
ok  	github.com/grafana/loki/pkg/chunkenc	31.598s

Compression speed benchmark

pkg: github.com/grafana/loki/pkg/chunkenc
BenchmarkWrite/none-16         	     754	   2003727 ns/op	 2758437 B/op	      48 allocs/op
BenchmarkWrite/gzip-16         	     106	  10503831 ns/op	  467140 B/op	      98 allocs/op
BenchmarkWrite/gzip-1-16       	     231	   5146374 ns/op	  448971 B/op	      98 allocs/op
BenchmarkWrite/lz4-16          	     427	   2878722 ns/op	  269449 B/op	      58 allocs/op
BenchmarkWrite/snappy-16       	     715	   1721530 ns/op	  750768 B/op	      56 allocs/op
BenchmarkWrite/snappyv2-16     	     296	   5321221 ns/op	10866139 B/op	     158 allocs/op
PASS
ok  	github.com/grafana/loki/pkg/chunkenc	29.111s

[owen-d]

Some new codepaths for me, but looks great.

[isavcic]

I see zstandard missing from this list, any particular reason?

[cyriltovena]

It was there, but the performance were bad so we removed it.