take control of exception object creation

[davidhewitt]

The idea of this PR is to move exception creation out of the Python interpreter (where we can't control it, e.g. all the special cases in #4412), into our own code.

I think this would pave the way for optimisations and maybe relates to #4584 and #4413.

Having started on this, I think for 0.23 it's going to be simpler to just introduce some locking inside PyErr, however I would like to revisit this in the future as I think it will be a positive step. So for now I'm just pushing to see impact on benchmarks & tests, and so I don't forget it.

[codspeed-hq]

CodSpeed Performance Report

Merging #4669 will degrade performances by 10.37%

_{Comparing davidhewitt:create-exception-objects (929056f) with main (f74d374)}

Summary

❌ 1 regressions
✅ 82 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	davidhewitt:create-exception-objects	Change
❌	err_new_restore_and_fetch	7.7 µs	8.6 µs	-10.37%