This factored computeTypeLocationSize() out of needing the TIntermediate contents,
and uses it to show how to know how many locations an object needs.
However, it still does not do cross stage, or mixed location/no-location
analysis.