Stats and logging improvements

[bhntr]

This is a first step toward two simultaneous goals

1. Make it easy to implement a parallel experiment runner for Popper
2. Make Popper easier to use programatically from within another
   Python application.

The major changes here are:

1. Configure the Popper loop to store information about its execution
   in the Stats object. The Stats object can then be serialized or analyzed
   programatically.
2. Add arguments to the Settings initializer so a Settings object can be
   created without argparse.
3. Perform all logging using python's logging framework rather than
   print(). This makes it easy for other applications to tune how the
   logging happens, set the log level, etc.

There should be no changes to command line Popper besides a very small
amount of extra memory usage.

For goal 1, future work will include an experiment runner in a
separate repo.

For goal 2, we probably want to add a setup.py file to build a .whl file
from Popper and publish it to PyPi.

[bhntr]

Ideally Settings would take strings for bias, examples and BK. This would make it a lot easier to just generate these programs in an experiment runner without having to create files. But pyswip makes loading a Prolog program from a string overly complicated. So the bias is a string and the BK and examples are files.

Taking these arguments separately (rather than as a folder path ala kbpath) makes it easier to load the files from different locations. In my experience, this can be convenient when running experiments which are minor variations on each other.