feat: better compression bandwidth

[mterenzi]

Change

nmsgtool -z compresses on the reader thread. While zlib runs, nothing drains the
socket. A single reader tops out before the kernel receive queue overflows and payloads are lost.
--zasync moves compression onto worker threads
and puts the writes back in order behind one committer thread.

Using it

--zasync n      up to n compressor threads (-1 chooses, 0 off — the default)
--zcull secs    give up a compressor idle this long (300; 0 never)
--zmin n        never cull below n compressors (1; 0 lets the pool empty)

Replace zlib with zlib-ng

default:              libz.so.1 => /usr/lib/x86_64-linux-gnu/libz.so.1
LD_LIBRARY_PATH=<ng>: libz.so.1 => .../zlib-ng-233/lib/libz.so.1

Related Confluence Page