Add io arguments to code_*, and improve reflection tests

[timholy]

code_native, code_llvm, and code_warntype display their results rather than returning an AST. This allows one to specify an IO argument to control where that output ends up. In turn, this allows one to write better tests for these functions.

For code_warntype, this also adds all-caps emphasis for when Base.have_color == false, and for good measure throws in a more careful disabling of emphasis when printing "raw" Expr objects.

@tkelman, I wonder if this might work around #9501? If so, it would suggest that redirect_stdout and redirect_stderr might have problems.

[Keno]

👍 LGTM, we have io arguments on everything else that prints, so I don't see why these should be special.

[ivarne]

+1 from me too. I would even argue that the AST returning versions also could accept a IO argument.

[timholy]

AFAICT, the AppVeyor failures are just timeouts.

[tkelman]

I sort of dislike the warnings indicate normal behavior: bit.

[timholy]

I know what you mean. Are you wishing we would suppress them altogether? I didn't want to do that given #9501.

[tkelman]

That's fair. If we can trace that down (more likely Jameson or Keno, though I must say that was an impressive bit of bughunting you managed so far) and come up with a solution we're confident in, then we can eventually suppress them. I am a fan of the silence is golden principle when it comes to tests, though also haven't worked too hard to suppress the known warnings from the tests on Windows. I know those have confused a few newcomers.