Performance TODOs

[jprochazk]

• Tracing garbage collector
• Switch to register-based bytecode (without accumulator)
• Fixed-width bytecode + use an enum for instruction encoding
• Improve loop codegen (put latch at the end of the body, eliminates 2 jumps)
• Rewrite VM from the ground up, thoroughly benchmarking it to ensure it is competitive in performance against other VMs in the same category
  - Lua 5.4
  - Python 3.11
  - Node.js --jitless