Add support for PR uploads

[tresf]

Adds support for leveraging upload artifacts for PRs. Targeted at the master branch only.

Justification and planning available in #4035; closes #4035.

Also removes Qt5 from the build system, closing #2611.

[tresf]

@zapashcanon FYI

[zapashcanon]

I'm wondering if there's another way to do it. Can't figure it out now, but will think about it.

[tresf]

I was thinking something like if [ foo=$(/usr/bin/true) ]; perhaps. Haven't tested it and it's a tad bit uglier.

i.e.

if [ response=$(curl -F "file=@$PACKAGE" "https://file.io") ]; then
# ...

[zapashcanon]

Haha, I tried it already, doesn't work. :p (Plus, there's some new warnings from SC because it thinks we want to compare response etc.)

[zapashcanon]

OK, actually, it seems to work, but, here's what I got from shellcheck:

 β ~ cat tmp 
#!/bin/bash

if [ response=$(curl -F "file=@$PACKAGE" "https://file.io") ]; then
	echo cc
fi


 λ ~ ./tmp 
Warning: setting file   failed!
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   145    0     0    0   145      0    145 --:--:-- --:--:-- --:--:--   508
curl: (26) read function returned funny value
cc

 λ ~ shellcheck tmp 

In tmp line 3:
if [ response=$(curl -F "file=@$PACKAGE" "https://file.io") ]; then
     ^-- SC2077: You need spaces around the comparison operator.
              ^-- SC2046: Quote this to prevent word splitting.

[tresf]

Thanks. Yes, that's worse. We'll keep it as-is for now, I guess. https://stackoverflow.com/a/6934931/3196753

[PhysSong]

In respond to failing LADSPA build... @lukas-w disabled C++11 for LADSPA in #4000 and it seems to cause build issues with clang. And -Woverloaded-virtual error seems able to be fixed by simply adding const qualifier.
By the way, why -Werror turned on in Mac build? It was disabled before and I can't find any relevant changes in this PR.

[tresf]

@PhysSong thanks, I think I'm going to just add the boolean logic back in. This seems like a mistake in #4000.

[lukas-w]

Yes that change was unintentional. Doesn't make sense changing the behaviour in #4000 as it doesn't build with MSVC either way. Sorry about that.

[tresf]

Commits squashed.

Note, as part of the Qt5-by-default changes, I've added a new cmake module QtLocator, which aids in fetching the Qt_DIR path for setting up a custom CMAKE_PREFIX_PATH (such as in a non-standard Qt install) but after the cmake process has been fired. It's boilerplate for homebrew/mac now, but can be extended to look for Qt on Windows, Linux as well as QtCreator bundled versions.

The build issues on Apple were due to a bug in script.sh, which overrode -DUSE_WERROR=ON. I've checked and this is explicitly set in win32, win64, linux and osx scripts, so this has been removed from script.sh.

[tresf]

• Re-squashed after rebasing against all changes against stable-1.2 (thanks @PhysSong).
• Per request on Discord QtLocator has been moved to a separate PR Add better Qt5 detection #4067.