Use a better compression algorithm for directory stuff

Have a look at these results:

[554]$ cat cached-microdesc-consensus |wc -c
  595929
[555]$ gzip -c -9 cached-microdesc-consensus |wc -c
  255689
[557]$ xz -c -9 cached-microdesc-consensus |wc -c
  238396
[556]$ bzip2 -c -9 cached-microdesc-consensus |wc -c
  222433

Now, we might still want to stick with zlib for general-purpose compression of descriptors and microdescriptors, since we need to compress those streams on the fly. But for consensus documents, which change infrequently, we don't much care about the CPU load required to compress them.

Sending 13% less bytes on consensus traffic would be a clear and win.