Commit 4cf70b14 was causing the tests to
crash.  The reason is due to a Wine bug in NdrPointerUnmarshall().
Windows returns the same pointer (which is later released in a call to
NdrPointerFree()), while Wine makes a copy which Valgrind correctly
spots is leaked.

Calling wined3d_texture_bind() + context_invalidate_active_texture() is
equivalent to just calling wined3d_texture_bind_and_dirtify().

I missed this in 7c4d512f. Generally speaking,
the only places where plain wined3d_texture_bind() calls are appropriate are
places where you're binding the texture to a specific stage for a subsequent
draw. I.e., places where context_active_texture() is called first.

wined3d: Bind the texture before calling wined3d_volume_upload_data() in wined3d_volume_load_location().