Try to Pool / re use compressors where possible. Fixes #2788

[abaranec]

No description provided.

[niloc132]

Looks good. Filed a follow-up issue #2920 for myself, unless you want to handle any of these points in this patch.

• Disable/fix meaningless "Got brand-new [de]compressor" messages. zstd, lzo, and gzip end up logging multiple times when decompressing each table, and zstd and gzip also multiple times when compressing.
• Once we find that we can't ask the pool for a compressor, quit asking. This will somewhat mitigate the above, but also keeps us from doing the extra work of asking the pool for an instance we know we can't find (which itself requires taking the lock, etc). Specifically, if innerCompressor is null and the compress method ends up calling compressionCodec.createOutputStream(os), we should keep track of that fact. This will specifically protect the zstd and gzip codecs from having to rediscover this for each page within each column, as well as for the dictionary for each column.
• For x86, investigate providing the native codec, which will enable the gzip codec not worry about the above case and pool its de/compressors.