We will need them in buffers for Vulkan, as well as for software
vertex shaders. Currently we simply store them in sysmem buffers and
read directly back out of them to load GL uniforms.

This does decrease memory usage a bit for d3d10+.

All other code touching stateblocks is in stateblock.c; this is a more intuitive
and modular place for it. This may merit renaming the function, but should be a
subsequent commit.

This syncs the code with dlls/ntdll/unix/env.c add_registry_environment.
It fixes ALLUSERSPROFILE and ProgramData environment variables value
in services.

I suspect the intent of this test was to verify that httpapi really doesn't bind
to that port. That HttpAddUrlToUrlGroup() fails implies this, but paranoia in a
case like this is rarely unwarranted.

However, this test cannot reliably succeed; some other operating system
component may open a TCP port in the meantime, and it appears that on Windows 10
this does in fact happen.

Since the goal of the test is just extra paranoia, I don't think it's worth
keeping.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55127

So that it is done atomically with retrieving events.

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=52474

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=54413

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=54546

We are testing that FD_CLOSE isn't generated, but Wine currently generates
FD_CLOSE anyway. However, sometimes Wine does not generate FD_CLOSE immediately,
causing the test to intermittently succeed inside the todo block.

This test is difficult to fix properly—it would involve either (arbitrary?)
buffering in the server, or support from the host OS.

However, we also don't want to make the Windows tests take longer, so add the
200 ms of leeway only on Wine.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54073

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=54073