ole32: Don't allocate the ole clipboard object in global memory - we're not going to directly expose it to other processes.

msi: Test and handle the case where the SourceList key does not exist and the PackageName value does not exist in MsiGetProductInfo.

msi: Factor out the MsiEnumPatchesEx tests and test the NULL szUserSid and an szUserSid that is different from the current user.

I missed one case when converting over to static GUIDs in a previous patch.
Pointed out by Andrew Talbot.

This allows ~<< to be properly matched.

rpcrt4: Fix a memory leak in union unmarshall functions by never passing fMustAlloc=TRUE into union_arm_unmarshall.

Previously it was only done for pointers.

_StubMsg.pfnFree should be used instead of NdrPointerFree for
conformant-sized strings since the buffer will have been allocated
using NdrAllocate.

Non-conformant strings do not need any freeing since they use a buffer
that is allocated on the stack.

It appears there are some calls to wined3d missing critical sections.

This should make us crash when trying to use the "currently active" context
after it has been destroyed, rather than messing around with freed memory.