Class types in containers: help option, known subclasses and set parsing; harden import path policy

[mauvilsa]

What does this PR do?

Follow-up to #959 with a few unrelated fixes on top. Four commits, each self-contained.

Class types nested in containers

• --*.help is now available for class types nested in a tuple, set, frozenset or mapping (e.g. dict[str, SomeBaseClass]), not only in a list.
• Known subclasses shown in the help now include subclasses of class types in any container, computed from a single get_class_types helper shared by help, extra help and shell completion.
• Fixed: shell completion of a --*.help option gave wrong choices when the class type was nested in a container or optional, e.g. builtins.NoneType for Optional[list[SomeBaseClass]].
• Fixed: set/frozenset of a class type failed to parse because subclass specs are not hashable. Such values are now kept as a list while parsing and become a set on instantiate.

Import path policy

• jsonargparse itself is denied by default and rejected in import_path_allowlist, since a value naming it could call set_parsing_settings and change the policy that is checking it. ActionParser no longer goes through import_object for its own class.
• Extended the default denylist: builtins.breakpoint/help/getattr/ setattr/delattr/vars/globals/exit/quit, cProfile, profile, doctest, unittest, _frozen_importlib*, pkg_resources, ensurepip, _sitebuiltins, resource, faulthandler, codecs.open, winreg, xml, antigravity.
• Documented that type hints also bound what a config can reach, how objects without a defining path are attributed, and that the omegaconf parser modes bypass the denylist through OmegaConf resolvers (e.g. ${oc.env:...}) and should be avoided for untrusted configs.

Postponed annotations

• Literal with arguments that are attributes of the class in whose body the method is defined now resolves. The class attribute values are only added as aliases for names inside a Literal subscript, so they can't shadow a global in a type position.

Misc

• Removed the last pragma: no cover in _stubs_resolver.py, covered by a test.
• The without-future-annotations tox env now uses {envtmpdir} instead of a fixed /tmp path, and sed -i.bak so it works on macOS.

Before submitting

• Did you read the contributing guideline?
• If you used a coding agent, did you fully understand and validate all generated code and ensure it follows the contributing guidelines?
• Did you update the documentation? (readme and public docstrings)
• Did you write unit tests such that there is 100% coverage on related code? (required for bug fixes and new features)
• Did you verify that new and existing tests pass locally?
• If this is a bug fix, did you verify that the tests fail without the code fix?
• Did you make sure that all changes preserve backward compatibility?
• Did you update the CHANGELOG including a pull request link? (not for typos, docs, test updates, or minor internal changes/refactors)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (09f2428) to head (59fcfd6).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #960   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           27        27           
  Lines         8762      8784   +22     
=========================================
+ Hits          8762      8784   +22     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud