Very early on, we wrote unit tests for the perforce implementation that

connected to public.perforce.com:1666. This worked okay for most things,

but it always felt kind of wrong.
This change makes it so we'll instead spin up a local p4d. This works

for almost all of our tests, and the slight delay I had to add during

that spin-up process is offset by no longer having to make actual

requests to the internet.
It seems like something in the Perforce tests was actually making a lot

of the Perforce tests run successfully even if they were broken. I'm not

sure which of my changes uncovered it, but there were several tests for

revision parsing which were obsolete (since revision parsing moved

completely into the diff parser). I've just removed those tests.
reviewboard/scmtools/testdata/p4_repo is part of the commit but has

been excluded from the posted change.

Ran all perforce tests and saw that all of them actually ran, and that

the previously broken stunnel tests now passed.