Xi2 build fix

[Simon Thum]

Peter Hutterer wrote:
> this code hasn't been touched for ages (it's even disabled by default).
> hal.c came after this code, so the use of RemoveDevice is more legacy than
> anything else.
> 
> Have you tried DIDR? Does it work fine? If so I'll just amend your patch
> before pushing.
Not really tried it. It's a in recovery path anyway, so to test I would have
to provoke error conditions somehow, which in turn doesn't really count 
as a realistic
test.

Looking at the xf86 impl, we really should be calling DIDR instead. The 
kdrive impl needs a RD(..., TRUE) fix too BTW. Others are stubs.

I'm all for DIDR, since otherwise the xf86 path leaks resources.

Cheers,

Simon