Implements parallel processing of satobs

[interplanetarychris]

Where multiple processors are available.

About 2x faster on a 2-core system, and 3-4x faster on an 8-core system A balance between regular user feedback and maximum usage of CPUs results in less-than-fastest possible performance.

For fastest performance, increase chunk size to a value larger than cpu_count. (Perhaps make this a parameter?)

Currently uses os.cpu_count() to determine number of threads available, which may be "too much" in some circumstances. Rather than make a dependency on psutil, it might be better to create configuration options for number of threads, and batch size in a future update.

process.py

• Implements a multiprocess worker pool of size os.cpu_count()
  • Batches processes in size os.cpu_count()
  • Creates pixel_catalog file using source FITS filename, instead of "test.cat" - Forces single-threaded astropy IERS table update, if needed
  • Receives all file and screen output from children, so that it can be output in the order of the files processed
  • Changes display order of FITS file being processed (first) followed by SATIDs within
  • Adds temporary notice of empty input queue
  • Adds description of screen color codes

astrometry.py

• Adds fall-back method of shutil.copyfile if PermissionError from copy2

extract.py

• Removes terminal output, and instead returns results to calling processes
• Tightens up file/io loops

satellite.py

• Saves hough3dlines output to unique tmp file instead of /tmp/hough.dat
• Removes temp file after processing
• Adds FIXME note to review for unnecessary code

stars.py

• Generates star catalog directly to destination file
• Updates sextractor call to reference destination instead of tmp file

[cbassa]

Great work on parallelizing this part of the code! Making the processing steps thread safe is going to be very useful. We should probably have a discussion with @ppapadeas on modularizing the processing steps, as this is also something I have been thinking about. In that sense it could be an option to parallelize over tasks instead of input files; i.e. one thread for astrometry, one thread for computing predictions, etc.