These arrays are both keyed by CPU name and go into the same tablegenerated file. Merge them so we only need to store keys once.
This also removes a weird space saving quirk where we used the ProcDesc.size() to create to build an ArrayRef for ProcSched.