gh-118761: Optimise import time for string

[AA-Turner]

This PR achieves a 27x improvement in import time for the string module. The main improvement comes from replacing Template.__init_subclass__() (GH-16256) with a descriptor class, allowing lazy import of the re module.

Current:

import string: cumulative time
mean: 9162.100 µs
median: 9133.000 µs
stdev: 66.662
min: 9071
max: 9301

This PR:

import string: cumulative time
mean: 334.967 µs
median: 329.000 µs
stdev: 13.438
min: 316
max: 368

• Issue: Improve import time of various stdlib modules #118761