For example, interpolating palette indices doesn't have the desired result.
Should we really want filtering for these cases we could implement it inside
the relevant shaders, after the fixup, but I doubt it's worth the effort.

Partly reverts my commit 3e5c704e.

shell32: SHCreatePropSheetExtArrayEx should check if the key itself contains the clsid of the shell extension.

Test %VAR%VAR% to check that the second variable is not expanded.

%CD% seems not supported on Windows NT 4.

Windows returns 32 byte aligned pointers when locking vertex and index
buffers, and some applications(Half Life 2, Alpha Prime, possibly others)
rely on this. Check the alignment and fall back to double buffered
buffers with HeapAlloced and aligned pointers if the alignment doesn't
fit.

This is needed to take advantage of asynchronous uploads with double buffered buffers

Shaiya locks a non-dynamic buffer with the DISCARD flag and expects
the contents to be retained. The SDK says DISCARD requires dynamic
resources, and Windows 7 returns an error in this situation, crashing
Shaiya. This patch sticks to the Windows XP behavior and allows the
lock, but ignores the DISCARD flag to retain the buffer contents.

dbghelp: Added full support for DW_FORM_data8 attributes, as well as primary support for DW_FORM_block* attributes.

This code was trying to block negative values of memid (incorrectly).
I have run more tests and found that it's possible to set some negative
values (e.g. on TKIND_INTERFACE: -1, -4..-8, -500..?). Unless we know
what values are allowed it's probably better to not check memid.