Compiler codegen: small refactor with progress fix

[ysbaddaden]

A small refactor that extracts a few methods from #codegen_many_units:

• #sequential_codegen(units)
• #parallel_codegen(units, n_threads)
• #wants_stats_or_progress?

Also fixes the behavior of stats vs progress: we only need to collect reused modules for compilation stats (not for progress). Sequential compilation now correctly increments progress in sequential mode (progress was stuck at 0 when threads=1 which is always true on Windows). Extracted as #14763.

TODO:

• [x] wait for #14762;
• [ ] wait for #14763;
• [ ] consider simplifying #codegen since it doesn't need to take care of the units.size == 1 case anymore; maybe even merge #codegen_many_units into #codegen :thinking:

Extracted from #14748.

[ysbaddaden]

AArch64 CI is failing with "fork not implemented" :raised_eyebrow:

EDIT: the issue is the legacy crystal compiler in the images (1.0.0 by default :scream:) that doesn't support the has_method? macro outside of methods, so we can't generate different method implementations :face_exhaling:

[straight-shoota]

Actually, I prefer the version with a single def and different bodies, instead of different defs.