Fix saltclass literal '^' list-override marker for single class

[ggiesen]

What does this PR do?

Fixes saltclass leaving a literal ^ element in the merged pillar when a list override is defined by a single class and no existing list is present to override.

In salt/utils/saltclass.py, dict_merge only stripped the leading ^ override marker inside the key-present list+list branch. When the key was absent from the target, control fell to else: a[key] = b[key], which plain-assigned the value verbatim. In real usage the override list is nested (pillars dict -> pkgs list), so the absent key is actually the pillars dict; the whole subtree was assigned by reference and the code never descended to the list, so the ^ survived.

The key-absent branch now descends into dicts (processing nested override lists at any depth) and honours a leading ^ on a list. An empty-list guard avoids an IndexError, and the marker is stripped with non-mutating slicing.

What issues does this PR fix or reference?

Fixes #50755

Previous Behavior

A class defining only an override list, for example:

pillars:
  pkgs:
    - ^
    - three

produced pkgs == ['^', 'three'], so the literal ^ leaked into state runs (e.g. The following packages would be installed/updated: ^).

New Behavior

The same class produces pkgs == ['three']; the ^ marker is stripped whether or not a prior list exists to override. Values with no ^ marker are unchanged.

Merge requirements satisfied?

• Docs: N/A (no user-facing docs change; behaviour now matches documented override semantics)
• Changelog entry (changelog/50755.fixed.md)
• Tests written and passing

Tests added in tests/pytests/unit/utils/test_saltclass.py: a direct-altitude test using the production-exact nested shape from expanded_dict_from_minion, a deeper-nesting variant, an inverse test proving the existing key-present override path is unchanged, a marker-free key-absent test, an empty-list guard test, and a plain-extend test.

Commits signed with GPG?: No