Using GL_NV_vertex_program2_option so far. If we're really desparate we can
handle some cases without the extension by using a custom varying and texkill
in the fragment program.

The view matrix that was set at the time the clipplane was applied is
hardcoded into the clipplane. With vertex shaders clip planes ignore
the view matrix in d3d. We implement that, but this also means that
the clip planes potentially change when we switch between vs and fixed
func.

This gives a small performance improvement. Don't enable NVfp for it though,
because the NVfp penalty is bigger than the gain from this patch. But if NVfp
is enabled anyway, make use of it.

The ARB "LRP" instruction is a fragment program only instruction. It is
however valid in vs_2_0+ vertex shaders, so we have to emulate it.

riched20: If DataObjectImpl_EnumFormatEtc returns E_NOTIMPL then ensure that returned pointers are nulled.

The write mask sizes for the input and output semantics can be different,
resulting in compilation failure.

ole32: In simple create mode the size returned by IStream_Stat is actually the current stream position.

This avoids allocating them on the heap, and avoids the need for a
separate vtbl.

rpcrt4: Return the stack size from the ObjectStubless function so that the thunks are independent from the number of arguments.