Avoiding compilation overhead in fragments where compilation is doomed to fail

It would be great if PaSh could determine (using profiling) that a fragment that failed can never succeed in a different loop iteration, for example if the fragment:
- uses a non-pure command, then there is no way it will succeed in a later iteration
- does not depend on a file (e.g., just `echo`) and takes a small time/cannot be parallelized even if pure
- ...

Adding new conditions should be pretty simple, the main complexity is designing the mechanism that avoids going to PaSh mode.

Then it could set an environment variable to completely avoid going to pash mode when in that fragment. This could reduce PaSh's overhead significantly, especially in loops that are the biggest source of PaSh overhead at the moment.

The issue names no files, tests, or entry points. Start by locating fragment retry behavior and the mechanism that enters PaSh mode, then investigate how profiling could record fragments that cannot succeed in later loop iterations. Done means qualifying fragments set an environment variable or equivalent state that avoids PaSh mode and reduces loop overhead, with coverage for non-pure and non-parallelizable cases.

Written by the indexing model from the issue text.