Specify the encoding and enable warnings in tests when unspecified.

[jaraco]

• Enable PYTHON_WARN_DEFAULT_ENCODING to catch EncodingWarnings. Ref EncodingWarnings in utils.subprocess and cache modules #12070.
• Specify encoding in Popen. Fixes EncodingWarnings in utils.subprocess and cache modules #12070.

[jaraco]

I'm unsure if the test is correct here. When I ran nox on the project with the test commit, it didn't emit any warnings. It was unclear to me if that was because I wasn't using nox correctly or if maybe the broken behavior isn't triggered as part of the run or if I've implemented the environment variable incorrectly. I'd appreciate if someone would take over, validate the test (and amend it to work), and add the required changelog history. Thanks.

[pfmoore]

What's the justification for using UTF-8? Can we be sure that any tools that we call via the call_subprocess function will output UTF-8?

I know, for example, that we pass the setuptools output through that interface. And I also know that (at least in the past, it may be fixed now) MSVC didn't use UTF-8 for its output. Does setuptools now force tool output to be UTF-8, or do we get the raw output bytes from build tools?

If the logic is "even if UTF-8 is wrong, we will only get mojibake, things won't fail" then we should be explicit that this is what we're doing, and confirm we're OK with it. Personally, I can be persuaded it's the best choice, but I expect we'll get user complaints about "pip garbling the output of my compiler", and I'm a bit fed up of pip being blamed for things like that...