Honor interpreter constraints even when PEX_PYTHON and PEX_PYTHON_PATH not set

[blorente]

Solves issue #656, where if no PEX_PYTHON_PATH and no PEX_PYHON are configured, we still honour --interpreter-constraint flag when bootstrapping a PEX.

There is an integration test that worked only when applied the last commit, but there is a lot of nuance to running it that I'd like a hand with. In particular, I'd like to know if there's a way of:

• Specifying "please ignore system-wide settings for PEX_PYTHON_PATH" in a test or Specifying
  "please override PEX_PYTHON_PATH with None". I think the first one should be easy to implement with a command-line flag defined in the same place as --rcfile.
• Specifying "This test only runs on Python3.5", because I need two different interpreters to play with (one for the environment that the pex runs in, another one to pass to --interpreter-constraint).

[jsirois]

@blorente for point #2 you want this:
https://github.com/pantsbuild/pex/blob/edc636ce6a004fa1bfc762d74f01c9fd3a862429/pex/testing.py#L324-L366

For point #1 you should just be able to run pex with an env=... that has what you want. And part of what you want is maybe:
https://github.com/pantsbuild/pex/blob/edc636ce6a004fa1bfc762d74f01c9fd3a862429/pex/variables.py#L322-L328

[blorente]

Depends on #673

[blorente]

For some reason I can repro failing test (and others) locally in master, even after blowing up ./.tox. I have no idea what's up or why they don't show up in other places. I will keep investigating further, but ideas are welcome.

[blorente]

@jsirois, any idea why I can repro the failing tests in master? Meaning, how they made it past CI in the first place.

[Eric-Arellano]

Thank you Borja for working through this!

[Eric-Arellano]

@jsirois or @CMLivingston a review of this would be quite helpful. While I'm more familiar with Pex than a few months ago, I'm still not aware of all the expectations and its history.

[CMLivingston]

LGTM - no issues that I can see.

[CMLivingston]

@jsirois, any idea why I can repro the failing tests in master? Meaning, how they made it past CI in the first place.

In my experience, this happens because certain env conditions align locally that are not covered in CI, and the developer did not verify his/her changes by running tox locally before pushing the branch.