This restores the functionality of "clicking one-time cc checkboxes
adds them to the textarea."  It was previously falling back to the
non-JS version.

The empty box, while not incorrect, confuses end-users.  This also
makes the Reply page mirror the functionality that is seen on the
Basics and People pages.

It is possible for the server to have sent us the ready signal before
the parent hits the sleep(); in this case, the parent would wait,
uninterrupted by SIGUSR1, for 15s, despite the server being ready.

Hence, only sleep if we have not already received the signal.  There
is still a minute race condition between the "unless" check and the
"sleep," but this reduces the window considerably, and the side
effects to the race condition are merely a slower test.

Under high load, such as heavily concurrent testing, there is no
guarantee that the parent process of the fork will have time to set up
the signal handler before the shild finishes initialization.  Thus,
receiving a USR1 signal causes it to die, leaving a zombie child
process and the test wedged.

Set up the signal handler before we fork, which guarantees that the
child will not be able to force this race condition.  It leaves forked
server processes with a USR1 signal handler, but there is little harm
to this.

Create.html actually looks for Queue not queue

FindAllRows searches for disabled rows and is used quite extensively in
RT, so we can't just "fix" it (it was added in c723178c)

Status != Starts; the Starts field in question was being added by a
callback.

This reverts commit 4195dd1d.

I've been asked this question way too many times this week

This didn't come up previously because we relied on the ->Owner value
on the original ticket object to remain unchanged for purposes of
right checking.  The new rights SQL ACL query works around this,
however -- thus, we delay the untake until after the queue move.

->DryRun returns undef for such users; filter them out

This is only necessary because commits were mistakenly made by me on
the releng branch.

To avoid causing engineer strife, I'll let someone doing release
engineering actually _run_ the tagger

Skip a bunch of invalid SQL warnings on invalid TicketSQL by overriding it with a known-good empty result set