The OF code for parsing bus-width would only add the specified width
to those the driver might have already set capability flags for.

Because of this, if the driver had set 8 or 4 bit width, it wasn't
possible for the DT to specify that fewer pins were used on the board
and a smaller width was necessary.

Change this so the width in the DT overrides whatever widths the
driver says it supports.  There is no reason to have an incorrect
device tree and it makes far more sense for the DT to override the
driver default than for the driver default to override the DT.

The widths the driver puts in host_caps before calling mci_of_parse()
are considered the default if the DT doesn't specify bus-width.  This
should cause the least amount of change to existing boards, as despite
a comment that no bus-width meant to use 1 bit, using the driver
default is what was really happening.

Unfortunately, half of existing drivers default to the largest width
they support while the other half default to the smallest.  Boards
should just stick the width in the device tree.

Signed-off-by: Trent Piepho <tpiepho@kymetacorp.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>