Refactor stats caching and eliminate binary format

[jqnatividad]

When the stats cache was first implemented, we created a binary format of the stats in addition to the CSV cache, with the --stats-binout option.

This was an attempt to directly load the stats into memory, directly into the Stats data structure and eliminate parsing.

However, it seems the decompression/direct-load approach actually takes a long time. Even longer than parsing the CSV file!

For "automagical" commands that uses Stats caches, we should just parse the human-readable CSV version of the cache, and eliminate the binary format.