More prep work for new opcodes.

The opcode-gen script now expects there to be up to 65536 opcodes, and
it generates opcode chain information. The chaining info isn't used
yet, but eventually it will get hooked into the instruction selection
logic in dx, meaning that dx will no longer have to do relatively
slower opcode searches; the right "next" opcode will have been
staticly generated and compiled in.

I also added a helpful generated comment in RopToDop, which indicates
which opcodes are the ones that *won't* be found by chaining, that is,
which ones are the starts of chains. This is left as a comment and not
code, though, since there are lots of quirks with how these get hooked
up. I hope to make this cleaner at some point, but today's not the day
for it.

Change-Id: I11da833cc57b383733743114c0f3e0e64903982d