CFG Part 0: Max Stack Size & Peephole Optimizer Fixes/Improvements

[ike709]

This PR does a few things:

• Steals some redundant Jump optimizations from Control Flow Graph #2558
• Fixes an issue with a case in which we did not rescan opcodes after calling AttemptCurrentOpt()
• Changes the peephole optimizer to keep rerunning until each pass reports no changes to the bytecode. This provides further gains in some operations involving stack pushes and list creations, and generally makes it less annoying trying to spread the optimizations out across separate passes in the correct order. This could also lead to further simplification of the peephole optimizer in the future, but I didn't want to change too much all at once.
• Fixes some incorrect stack size/delta info
• Fixes max stack size not being recomputed after optimizations. This leads to reductions in stack size across thousands of procs.
• switch() cases no longer emit a Jump instruction if the last operation is a terminal change in control flow (e.g. return)
• spawn() blocks no longer emit PushNull and Return if the last operation is a terminal change in control flow (e.g. return)
• Peephole optimizer now ignores AnnotatedBytecodeVariable during opt scanning
• Pointless Jump instructions are more aggressively pruned by the bytecode optimizer
• Made some fixes to ProcDecoder that arose during testing

I did before & after diffs on /tg/station compilation & disassembly to confirm that this resulted in further optimizations to the bytecode without introducing any apparent regressions. The NullRef opt in particular was being left on the table in thousands of cases, while the other peephole optimizer changes are less dramatic but still present.

The fixes to max stack size are also a huge improvement. Thousands of procs have either had their max stack size reduced to 0 or have seen a reduction from double to single digits. I ran TG through round init to confirm nothing exploded.

This sufficiently beats the codegen and optimizer into submission to the point that we can now implement dead code detection (after CFG is finished) and run it on /tg/station and goonstation without getting false positives.

[ike709]

I'm calling this PR done barring any review comments. I did another testing pass without any of my other non-committed experimental changes, and confirmed that TG and goon both compile, while also making sure I can start a round on TG.