Pytestless doctests 2

[pp-mo]

Replaces #177
Rebased to resolve conflicts,
and, since we moved the repo into SciTools, also replace the branch with one in my fork (pp-mo/ncdata)

##TODO
Issues inherited from #177

• explain that 'private' does not apply to docs files
• find better way, or document, the odd way we are doing globs (update: it seems there just isn't an easier way)
• tiny docs fix
• explain limited translation of args for doctests function calls, avoiding exec
• additional remaining
  • now complicated enough that it needs some tests
  • possible problems with less usual glob forms (in include/target specs) (update: OK I think)
  • support proper globs in exclude patterns ? (update: investigated, not very useful + too hard to explain)
  • lack of global options config (? use a pyproject.toml section ?)(update: investigated, too complicated to integrate with standard argparse, since defaults (especially bools) obscure what was actually commanded)

(and some more added since)

• output reports currently say like e.g.
  1 item had failures:
17 of 30 in lenient_metadata.rst
***Test Failed*** 17 failures and 9 skipped tests.
  • this should be clearer + especially include count of successes, so : "ok/bad/skip add up to total"
• ability to detect + skip internal sections that APPEAR like doctests, e.g. in ".. codeblock::"
  (dropped: non-trival and would require deeper engagement with the doctests api, beyond 'testmod/testfile)

[pp-mo]

status update 2026-08-20

current version has simplifications to the runner
tests okay-->fail for python 3.10,3.11 --> 3.12/3.13
= iris 3.14 --> iris3.16 (needing set_auto_chartostring support)

Now ticks all the above to-do boxes.
I think this code is now "good-to-go" as far as the doctest-runner and its tests are concerned,
but this branch is now a mess : a bunch of unrelated commits mixed in.

Will get the tests+CI fixed, and then re-propose this code
- possibly also using scitools templating ?