Presently there are a few places that make assumptions about the
existence of SCFCR, which doesn't hold true for several port types. While
generally harmless, this does lead to bogus reads/writes in both the
termios/runtime PM cases that are better off simply never being made in
the first place.

While we're at it, also get rid of a straggling PORT_SCI check that
infers all non-SCI ports contain SCFCR. This doesn't presently have any
impact, but as we're now able to test for the existence of registers
without defering to the port type we future proof for additional port
types.
Signed-off-by: NPaul Mundt <lethal@linux-sh.org>