Executing instructions on the emulated CPU currently consumes 0 time in a Shadow simulation. This can lead to inaccuracy in applications that are CPU-intensive.

Ideally we'd like to model CPU time in a way that preserves determinism, but optional features that model CPU time at the cost of determinism would also be helpful.

This is related, but mostly tangential, to avoiding deadlock in busy loops. See #1792 for busy loops in general, and #2066 for native preemption to get out of "no syscall" busy loops.