Buffered compression column by column for native protocol

[gingerwizard]

closes #755

This is probably sufficient for now. Ultimately we should handle this better as part of #809

[gingerwizard]

Few pending things:

• We need to decide what size we should flush on i.e. MaxCompressionBuffer. The original issue Compress insertion data column by column to reduce memory usage #755 used 1Mb. This feels too low IMO @genzgd thoughts?
• Needs docs inc. recommendation on how to reduce memory usage (we should document block buffer wrt to this as well). Document MaxCompressionBuffer.
• We need to benchmark and confirm reduced memory + impact on performance. Initial tests suggest this is actually faster.
• Add some more tests around this - forcing compression.

[jkaflik]

@gingerwizard @genzgd

We need to decide what size we should flush on i.e. MaxCompressionBuffer. The original issue #755 used 1Mb. This feels too low IMO @genzgd thoughts?

I found max_insert_block_size is 1048449 bytes default, so IMO it makes sense to keep this value since we flush every MaxCompressionBuffer is met.

[jkaflik]

benchmark/v2/write-compress-buffer-limit/write_test.go benchmark against main without MaxCompressionBuffer:

BenchmarkWrite1KB
Alloc = 925 MiB	TotalAlloc = 4024 MiB	Sys = 1920 MiB	NumGC = 38
BenchmarkWrite1KB-10      	       1	2201197625 ns/op
BenchmarkWrite16KB
Alloc = 1153 MiB	TotalAlloc = 8047 MiB	Sys = 1920 MiB	NumGC = 78
BenchmarkWrite16KB-10     	       1	2093274875 ns/op
BenchmarkWrite64KB
Alloc = 925 MiB	TotalAlloc = 12069 MiB	Sys = 2202 MiB	NumGC = 118
BenchmarkWrite64KB-10     	       1	2067872083 ns/op
BenchmarkWrite256KB
Alloc = 1153 MiB	TotalAlloc = 16092 MiB	Sys = 2203 MiB	NumGC = 158
BenchmarkWrite256KB-10    	       1	2066669625 ns/op
BenchmarkWrite512KB
Alloc = 925 MiB	TotalAlloc = 20115 MiB	Sys = 2203 MiB	NumGC = 198
BenchmarkWrite512KB-10    	       1	2065087792 ns/op
BenchmarkWrite1MB
Alloc = 1153 MiB	TotalAlloc = 24138 MiB	Sys = 2203 MiB	NumGC = 238
BenchmarkWrite1MB-10      	       1	2047471958 ns/op
BenchmarkWrite5MB
Alloc = 925 MiB	TotalAlloc = 28161 MiB	Sys = 2203 MiB	NumGC = 278
BenchmarkWrite5MB-10      	       1	2059622667 ns/op
BenchmarkWrite10MB
Alloc = 1153 MiB	TotalAlloc = 32184 MiB	Sys = 2203 MiB	NumGC = 319
BenchmarkWrite10MB-10     	       1	2039583250 ns/op

benchmark/v2/write-compress-buffer-limit/write_test.go benchmark against compress_by_column with various MaxCompressionBuffer values:

BenchmarkWrite1KB
Alloc = 686 MiB	TotalAlloc = 3229 MiB	Sys = 1459 MiB	NumGC = 37
BenchmarkWrite1KB-10      	       1	1973350542 ns/op
BenchmarkWrite16KB
Alloc = 802 MiB	TotalAlloc = 6458 MiB	Sys = 1459 MiB	NumGC = 77
BenchmarkWrite16KB-10     	       1	1883714208 ns/op
BenchmarkWrite64KB
Alloc = 686 MiB	TotalAlloc = 9687 MiB	Sys = 1459 MiB	NumGC = 118
BenchmarkWrite64KB-10     	       1	1920653250 ns/op
BenchmarkWrite256KB
Alloc = 895 MiB	TotalAlloc = 12915 MiB	Sys = 1459 MiB	NumGC = 158
BenchmarkWrite256KB-10    	       1	1867119792 ns/op
BenchmarkWrite512KB
Alloc = 686 MiB	TotalAlloc = 16144 MiB	Sys = 1459 MiB	NumGC = 198
BenchmarkWrite512KB-10    	       1	1871656875 ns/op
BenchmarkWrite1MB
Alloc = 686 MiB	TotalAlloc = 19372 MiB	Sys = 1460 MiB	NumGC = 237
BenchmarkWrite1MB-10      	       1	1864227708 ns/op
BenchmarkWrite5MB
Alloc = 895 MiB	TotalAlloc = 22601 MiB	Sys = 1460 MiB	NumGC = 277
BenchmarkWrite5MB-10      	       1	1879189750 ns/op
BenchmarkWrite10MB
Alloc = 686 MiB	TotalAlloc = 25829 MiB	Sys = 1460 MiB	NumGC = 317
BenchmarkWrite10MB-10     	       1	1894366416 ns/op

---

there are visible throughput and memory allocation improvements.