Remove temp build dir even if the req is not installed (fixes #420)

[claymation]

This change ensures that RequirementSet.cleanup_files() is called whenever req_to_install._temp_build_dir is defined, preventing detritus from accumulating in $TMPDIR.

[travisbot]

This pull request fails (merged 48d1d87c into c6789f6).

[claymation]

Note that the travisbot failure above seems unrelated to the changes in this pull request. It seems like an infrastructure failure. Can you try to build this pull request again?

[hltbra]

Just to reinforce what @claymation says I got it from travis-ci:

Executing your before_install (sudo apt-get install subversion bzr mercurial)
took longer than 900 seconds and was terminated. Consider rewriting your stuff
in AssemblyScript, we've heard it handles Web Scale™

[qwcode]

I think @pnasrat is the admin on our travis account, but I've seen other people make a trivial change (like change a commit timestamp or a commit comment), and then force push to re-trigger a build when it's obviously not the pull request's fault.

[qwcode]

it could be helpful to mention that the mechanism is that ProcResult.assert_no_temp gets called and throws an exception if the tmp is non-empty.
that will make people feel better that this test can actually fail.

[claymation]

Ahh, sure, feel free to add the gory details if you'd like :)

[qwcode]

the point is to be clear that the "mechanism" can actually cause a test failure, since you're not doing an explicit assert.

[claymation]

Thanks, I'll force push to trigger a build.

[travisbot]

This pull request passes (merged 865af46 into c6789f6).

[qwcode]

I want to merge this in, but I'm heading on a trip. I'll get back to this next week, unless somebody else merges first. I do want to fix that test.

[qwcode]

btw, I did also manually verify the "pip-random-build" temp dirs are getting cleaned up in this branch for the scenario mentioned in #420.