Stresstest tweaks

[rknop]

Changes include:

• util/logger.py adds a SCLogger class. Use that class in place of _logger everywhere, remove _logger from models/base.py
• Add a calibfile_downloadlock table to allow locking of calibfiles for a single instrument/section/filter so that two processes won't try to import them at once, and use that table in instruments.py. This is a temporarily solution, as it's just one examle of a bigger problem that we plan to solve a different way, but it's there for now as I needed it during the stress test.
• Add mpich to the Dockerfile. (Ended up not using it, but did use MPI for a while during the buildup to the stress test to see if it behaved differently then python multiprocessing for some things.)
• Everything in hacks/rknop ; these are the things I wrote to actually do the stress test. No tests written for these; the "hacks" directory should tell you that.
• In alignment, futz around a bit with swarp virtual memory (file paths and limits).
• In improc/scamp.py, add the ability for max_sources_to_use to be a list; if it's a list instead of an integer, iterate through it in an attempt to get a good solution. (This was needed to get successful astrometric solutions for a handful of the chips.)
• In models/image.py add set_corners_from_header_wcs. Also now call this during the overscan/trim part of preprocessing, and at the end of astro_cal if the image is flagged as not having a first astrometric calibration done.
• A commented-out function in provenance.py that tried to solve the database save race condition. (Didn't work, because it lead to lazy load failures with objects not properly attached to sessions.)
• In data_store.py, add an _overlap_frac method that is now used in get_reference. get_reference defaults to looking for the same instrument and filter, with a 85% overlap of the search image. It can optionally filter on target name and sensor section.
• Some Spin yaml files for my development purposes

[rknop]

Re: cleaning up the hacks : I've found it's often useful to have a git-tracked directory for individual users to put their experiments (things like what I did for stress tests). As long as we're in development mode, we should have these, but we might want to make a "make version 1.0.0" issue that cleans these out.

(On the other hand: realistically, most scientific pipeline code like this never really goes out of development mode, as the people running it are often the same as the people developing it. The stuff that I'm doing for the Roman space telescope supernova PIT ("project infrastructure team") is different, as eventually we're supposed to deliver stuff to another group for operations, but that's on a whole different scale.)