When opening the ucm* file, we need to select the correct file based
on the underlying libibverbs device. We cannot base the file name
on the uverbs* file name, since iwarp devices do not result in ucm*
files. Instead, we need to search for the correct file name based
on the device name.
This fixes a problem reported by Jeff Squires of Cisco running ib_cm
on a system with both iwarp and ib devices present.