Partial `reg` semantics for child nodes of indirect buses

[mbolivar-nordic]

This is similar to #170, except it's a situation where a
node has a reg property, but critical information is missing.

As of 4fc81d4, it's not clear what
happens if I map in a node via an address-map property which has one
of #address-cells or #size-cells set to zero in the parent node.
In this case, the node's register blocks are missing information that
we need to decide if the node is visible to the CPU cluster containing
the address-map property or not.

What do we do here?

• always visible?
• never visible?
• visible if the base address is visible when #size-cells is 0?

I guess a zero #address-cells but a nonzero #size-cells is not
practical, but I also don't know that it's explicitly forbidden by
DTSpec.

[sstabellini]

I think we should bundle this case together with #170 and say that if a child of a bus doesn't have a reg property (or #address-cells and/or #size-cells is zero) then it gets mapped if the bus is mapped.