Add native system semaphore and Windows shared memory

[DomClark]

The primary goal here is to eliminate the Qt requirement for RemoteVstPlugin. This helps compatibility with Qt 6, which does not provide 32-bit binaries any more. It also helps with the more general effort to remove Qt from the core, and simplifies the CMake scripts for RemoteVstPlugin a bit. RemoteZynAddSubFx still needs Qt for XML processing, but that can be dealt with separately.

Changes:

• Remove Qt shared memory implementation.
• Add native Windows shared memory implementation.
• Add native system semaphore with POSIX and Windows implementations. (The POSIX version is currently unused, as a different IPC mechanism is used on POSIX systems that doesn't use semaphores. I have tested it on a different branch in the past, however.)
• Tidy up some of the existing shared memory code.
• Remove Qt from RemoteVstPlugin CMake scripts, and thus install only a single Qt version in Windows CI.
• Copy the fix from Fix generator expression in strip command #6762 into the RemoteVstPlugin CMake script, as I forgot to do so at the time.

[Veratil]

Nothing stands out. Haven't tested but LGTM.

[Rossmaxx]

Yes, finally. Don't know what to test or review for in this but I'm glad to help here.

[Rossmaxx]

A user on discord has stated that CPU usage has increased in the nightly builds when using VSTs compared to 1.2. Is this a regression from this change?

[DomClark]

I don't know - there's so little information about the problem and so many changes since 1.2 that there's no way to confidently answer with "yes" or "no". However, it seems unlikely to me, given that the only code in this PR that runs many times per second (in particular, acquiring and releasing semaphores) is very simple.

If they are automating VST parameters, then the likely cause is #5321, which is a known issue. That feature is difficult to implement efficiently with the current IPC system, which is why I'm working on #6511. Otherwise, we need more information and more careful investigation.