bytes.ToLower allocates a slice and since this func is in the
hot path for write, it creates a lot of garbage.

This switches to a case-insensitive sort func that does not allocate.

This converts the marshalTS func to not allocate during marhsalling
which doubles the throughput and removes excess memory allocated
that needs to be garbage collected.

benchmark                  old ns/op     new ns/op     delta
BenchmarkMarshalCSV-10     2117          1141          -46.10%

benchmark                  old allocs     new allocs     delta
BenchmarkMarshalCSV-10     20             0              -100.00%

benchmark                  old bytes     new bytes     delta
BenchmarkMarshalCSV-10     1548          0             -100.00%

Bettter conveys that this is a one-time operation.

This sacrifices a little bit of disk for faster compression time
and lower CPU util.

The rings account for a lot of allocations and heap space.  Re-using
them cuts memory usage and GC work for the runtime.

These were too low for larger number of clients and cause new ones
to be allocated frequently.