Results in significantly better performance when optimization
is not possible.

This is only needed because of something like async-connect
followed by select (I guess?) and next thing we know we've
sent "N" bytes and still haven't told ipcd about our timings.

We should do this earlier, for sure.

Not yet vetted on all the apps, but resolves at least apt-get.

This should be done properly later, but for now
this should get our experiments running.

Unknown impact on performance/latency, but
should help the false positive case.

Clang warns about this, gcc docs seem to confirm it.

I could've sworn at one point "-O4" indicated
cross-module optimization (-flto and such),
but not sure where that idea came from.

Prep for possibly sending timing data to ipcd earlier.

FIND_PAIR has a silly number of arguments now, heh O:).

Plumbing is all there, but ipcd doesn't actually compare times yet.

One side should have created the socket via accept(),
the other should not have.  Track this and verify it
during the pairing process.

Committing this change from last submission push,
not entirely sure why this was done or wasn't done before.

Re-running numbers presently, but so far I haven't
observed any differences in results.

Bah, thought this script (used as template for others)
already handled this, must've never quite
finished it :/.