improve precision of coverage reports by including coverage by subprocesses

[daira]

Currently, coverage reports underestimate covered code in some places (this particularly affects the code for starting nodes) because only Python code directly executed by coverage run is measured.

coverage has a feature to measure the code running in multiple processes and then combine the measurements:

​http://nedbatchelder.com/code/coverage/subprocess.html ​http://nedbatchelder.com/code/coverage/cmd.html#cmd-combining

To make this work, we need to:

• Run this early on when starting Tahoe processes (e.g. in src/allmydata/__init__.py​):

  try:
      from coverage import process_startup
      process_startup()
  except ImportError:
      pass
  

• Set the COVERAGE_PROCESS_START environment variable (which will be inherited by subprocesses) when running coverage from setup.py and the Makefile​.

• Pass the -p option to the coverage command line.

• Ensure that the .coverage.* files are being generated in the same directory.

• After coverage run, execute coverage combine.