Merge pdbq:planned-merges-mm into pdbq:main

Some backend changes to pdbq for the merge board preparations.

The main thrust of this branch was integration of upstream data that's
now being generated by the `collect-watch` worker into the `usmerges`
backend script. collect-watch was rewritten from Bash to python this
last cycle and received a number of improvements in the process,
including switching the output format from text to JSON. So usmerges
is updated to now consume this JSON instead of custom parsing text.

Before:
  https://paste.ubuntu.com/p/RcR4KTHDGm/

After:
  https://paste.ubuntu.com/p/bN35xkR2X4/

The second thrust of this branch drops much of the "predictive
scheduling" functionality from `planned-merges`. This had proven to be
mostly inaccurate (thanks COVID era) and of limited actual value in
practice. While the improvements to collect better upstream data helps
address part of the former problem, the latter problem still trumps.
Users prefer to have the 'no-merge-available' cards left to the side and
handled manually. So instead of trying to (incorrectly) guess when
these merges might come to be, efforts will focus on automating the
detection of when they DO show up. The server team has been effective
at processing these on the fly when they come up in the housekeeping
meetings, so we can rely on that workflow instead.

Beyond that is assorted cleanup work and bug fixes to allow the merge
board data to be generated.