big error classification refactor

[softminus]

properly handle the new peer categorizations:

• split poll_this_time_around into poll_peers_this_time_around for peers and poll_hash_this_time_around version for hash tests
• make both handle new return values of get_classification
• in the slow walker, call poll_peers_this_time_around to see if we poll for peers, and call poll_hash_this_time_around to see if we poll for hash

error classification from zebra's errors:

• better classify errors in hash_probe_inner (look at grep failed |awk -F "failed: " '{print $2}'|sort -u to see the possible errors) to determine where the hash probe failed
• change the return type of hash_probe_inner to more elegantly return where the hash probe failed

interface with PeerStats

• change hash_probe_and_update to handle hash_probe_inner's return value, and update the counters (and the new Option<SystemTime>) in PeerStats properly

[softminus]

We decided not to have different polling intervals for peers and hash tests, but besides that, this is all done and working