In ntoskrnl.exe, when a PnP device interface is registered, the
interface path casing is retained, with only the hardware ID being upper
case. Thus, this patch looks to align the behaviour between PnP and
rawinput, particularly for games that seem to rely on these two strings
being consistent for hotplug.

This updated behaviour also seems consistent with Windows.

msvcirt's PARENTSRC imports exception.c from msvcp90, but it uses a function
pointer for these functions, resulting in a type mismatch.
Signed-off-by: Gabriel Ivăncescu <gabrielopcode@gmail.com>

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=56078

Fixes certificate verification in Brawlhalla.

Mostly written by Paul Gofman.

This test show that WM_SETCURSOR gets handled recursively, so in process_mouse_message(),
accept_hardware_message() can be called after sending WM_SETCURSOR and no changes are needed there.

Previously, accept_hardware_message() is called after WH_CBT HCBT_CLICKSKIPPED hooks. So when
these hooks call PeekMessage(), they will be called recursively. Note that WH_MOUSE hooks do
get called recursively according to tests.

Previously, accept_hardware_message() is called after WH_KEYBOARD and WH_CBT HCBT_KEYSKIPPED hooks.
So when PeekMessage() gets called in WH_KEYBOARD and WH_CBT HCBT_KEYSKIPPED hooks, the hooks will be
called recursively because the message is still in the server message queue.

Fix Toad for Oracle F3 find next function not working properly because its WH_KEYBOARD hook gets
called too many times.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=56062