Fix QEMU Plugin and add Instruction Address and Size Data; Make Branch Predictor Model Command-Line Parameter

[andrej]

• Remove potentially failing assertion on instruction size, as moving to x86_64, we now have variable-width instruction encoding and the size of an MCInst may not be known a priori
• To address previous issue, add new metadata field for instruction size, pulled directly from the input source; that way we know the instruction size is accurate
• Add missing extern TypeId declarations to QEMU plugin; without those, I got any_cast assertion errors between the shared library and the main executable (see this discussion I started a while ago on the LLVM forum)
• Make branch predictor unit model selectable on the command line: None, Naive or Skylake
• Print some statistics on branch predictions at the end

With these changes, I'm now able to run a hello world program via QEMU through MCAD. I had to disable the instruction cache modeling (-enable-cache=false on the command line) to make this work, or it didn't seem to terminate (maybe the cache model is very slow?)

Interestingly, there seem to be many more branch mispredictions in this hello world example. Maybe it will be worth it to look into some more benchmarking of this after all.

The changes to the vivisect plugin with the new instruction size field should work, but I haven't tested them yet since we're probably not super interested in it any more now that we have QEMU.