Not just those belonging to earlier completed fences.

completed_fence_id names the most recent fence to be completed. In essence, this fixes an off-by-one error.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=53217

Not just those belonging to earlier completed command buffers.

completed_command_buffer_id names the most recent command buffer to be completed. In essence, this fixes an off-by-one error.

The integer overflow bug in RtlUniform has been fixed since Windows
Vista.  Synchronize Wine's version accordingly.

Today, test_RtlUniform() skips almost all of its tests on Windows Vista
or later.  The skipped tests only succeed in Windows XP, and fail on
Windows Vista or later.

This is because RtlUniform()'s behavior has changed, presumably due to a
bug fix shipped in Windows Vista.  It appears that RtlUniform, a linear
congruential generator, could overflow before computing the modulo in
WindoWs XP.  This is no longer the case in Windows Vista or later.

Meanwhile, we no longer support Windows XP behavior and thus do not test
for it regularly; therefore, the tests are obsolete as of 2023.

Remove obsolete tests that no longer work with any of the Windows
versions actively tested by WineHQ (as of 2023), and replace them with
updated tests that works in the Windows versions from Vista up to 10.

We do not expect normal CreateWindow() and GetDC() calls to ever fail.

WH_CALLWNDPROC hooks are called on every message, but we only need it
for the activation messages. WH_CBT hooks are called for a few events
only, which is much lighter, as we need to use a global hook to track
windows which belong to any other thread.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54431

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54431

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54431

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54150

When the format is NULL vsnprintf() always returns -1 so we keep
doubling the allocated memory until running out of memory.