Fixed LD_LIBRARY_PATH squashing

[sternj]

Because of how the current environment is checked, LD_LIBRARY_PATH and LD_PRELOAD are squashed if present, rather than prepended to. This commit prevents this.

Also preemptively fixed a potential issue where PYTHONMALLOC would not be properly squashed if present.

[sternj]

I think that what we'll need to do (rather soon) is escape all of the unescaped spaces in environment variables in redirect_python, if there are spaces in any environment variable that we don't mess with then they'll cause issues if the wrapper script is ever invoked.

[emeryberger]

Any reason not to do this in the current PR?

[sternj]

It's potentially a more far-reaching change, I can't think of any unintended effects but they may exist, especially given that it's happening at the invocation site for any subprocesses. What I have here is the minimal change to fix the issue at hand, but I'll see what the test suite says about the further-reaching solution.

[sternj]

Note that this doesn't help with Windows environment variables, which I do not know the semantics of

[emeryberger]

Scalene doesn't make specific use environment variables in Windows, should be fine, but should also be tested. Can we get Windows testing into the CI?

[sternj]

I don't see why it isn't possible! I'll add it to the list.