feat: add module support to JustfileFixture

[bluefing]

Adds module-aware recipe resolution to JustfileFixture. Recipes from nested modules are flattened by namepath (e.g. infra::deploy::up), so all existing methods work transparently with full module paths.

Closes #1

Changes

• _flattened — cached single-pass walk of the dump, returning both recipes and modules
• _recipes — now returns the flattened dict keyed by namepath (was root-only)
• _flatten — static recursive walk of the module tree
• all_modules — property returning discovered module paths
• Backwards compatible — justfiles without modules work unchanged

Tests

9 tests covering: namepath flattening, privacy across modules, dependencies, parameters, unknown namepath errors, module discovery, private recipe filtering, and standalone _flatten usage.

[semanticdiff-com]

Review changes with  

Changed Files

File	Status
src/pytest_just/fixture.py	8% smaller
tests/test_module_fixture.py	0% smaller

[Mjboothaus]

Great direction and this adds high-value capability for modular justfile layouts.

Assessment
• Alignment: High strategic value
• Risk: Low–Medium (correctness/integrity edge cases)

What looks good
• Useful module-aware recipe resolution.
• Strong coverage intent across flattening, module discovery, and lookups.

Points to consider before merge

1. Collision handling: can we add explicit detection/error on duplicate flattened namepaths?
2. Strictness: should malformed module payloads fail fast rather than being skipped?
3. Dependency semantics: should dependency APIs return local names, qualified namepaths, or both?
4. Naming consistency: PR description references all_modules, implementation uses module_namepaths - can we standardise wording.

[bluefing]

Collision handling: can we add explicit detection/error on duplicate flattened namepaths?

Just does not allow duplicates, if I duplicate a recipe the in the same module I get this error

❯ just                          
error: Recipe `down` first defined on line 90 is redefined on line 105
   ——▶ .just/docker/compose/mod.just:105:2
    │
105 │ @down interactive="" profile="" *$services="": _require-docker _require-compose
    │  ^^^^

---

Strictness: should malformed module payloads fail fast rather than being skipped?

Good question. I think you are right, if the dump is malformed we should fail fast.

---

Dependency semantics: should dependency APIs return local names, qualified namepaths, or both?

Needs some thought - not sure what the right answer is here.

---

Naming consistency: PR description references all_modules, implementation uses module_namepaths - can we standardise wording.

Well spotted - I asked casey to implement a function to return the current module, he settled on module_path() as he considers namepath to be an internal name.

casey/just#3260

recipes() in pytest-just returns objects so I didn't want to use modules() as we are surfacing a list of string. I think it probably makes sense to follow just and have modules_paths()

What do you think?