Lost Recovery is triggered after a small number of
duplicate ACKs - potentially even without any duplicate
at all, if a sufficiently large SACK block is present
in the ACK.
By recognizing that some data was already delivered
before loss recovery formally started, PRR will
more accurately send out segments immediately after
loss recovery is eventually engaged.