Create parent bus_dma tags to mark the bus as cache-coherent

[andrew]

Summary

This marks the PCIe drivers used on arm64 as cache-coherent when this is
the case. In the ThunxerX cases we hardcode them to be coherent, and
in the generic PCIe driver we check the dma-coherent FDT property to
find if they are coherent.

I have tested this on both Pass 1.1 and 2.0 ThunderX with extra KASSERTs
to check the dma tags are all marked as cache-coherent.