Decide whether letting the read bucket go negative is actually helping us
Proposal 182 points out that we let the read bucket go negative (by reading the rest of the TLS record, which was already actually read by openssl so we might as well use what it says), but then we don't let the corresponding write occur, so we trap the cells inside Tor until the write bucket can catch up.
If that is so, then the feature is not helping us. And it may even be hurting us when combined with circuit priorities (aka ewma), since we make decisions about cell ordering before we need to.
We should compare 'with' and 'without' to see if this intuition is right.
(This task is motivated by #4682 (moved).)