Add Tier 1 unit tests for pure-logic helpers in main.py

[berendt]

Summary

Closes #247 — Tier 1 of the #232 unit-test rollout: unit tests for the
pure-logic helpers in netbox_manager/main.py. These functions are
deterministic transformations of plain Python data or tiny attribute-bag
objects, so the tests need no live NetBox, ansible_runner, filesystem, or
full pynetbox mocks — at most a types.SimpleNamespace.

Builds on the #233/#234 foundation (pytest harness, tests/ tree,
test-requirements.txt, shared conftest.py, Zuul netbox-manager-unit-tests
job).

Change set (commit order)

1. Add shared device/interface factories to test conftest — make_device
   and make_interface fixtures producing SimpleNamespace attribute bags,
   documented for reuse by later tiers.
2. settings/role and generic helpers (test_helpers.py, groups 1+2) —
   get_node_roles, get_switch_roles, deep_merge, get_leading_number,
   get_resource_name, get_device_role_slug,
   find_device_names_in_structure, is_virtual_interface,
   _split_tasks_by_type.
3. task-filter helpers (test_task_filters.py, group 3) —
   should_skip_task_by_filter, extract_device_names_from_task,
   should_skip_task_by_device_filter.
4. autoconf helpers (test_autoconf_helpers.py, group 4) —
   get_autoconf_prefix, _extract_device_names_from_autoconf_task,
   _filter_tasks_by_device, _filter_tasks_by_type_by_device,
   _filter_portchannel_tasks_by_device.
5. loopback-gate helpers (test_loopback_gate.py, group 5) —
   has_sonic_hwsku_parameter, should_have_loopback_interface (including
   settings-driven role-list overrides).

92 new tests; pytest tests/unit now reports 117 passed.

Local verification

• pytest tests/unit — 117 passed
• flake8 tests/ netbox_manager/ — clean
• black --check tests/ netbox_manager/ — clean
• mypy netbox_manager/ — unchanged from baseline (pre-existing
  import-untyped notes for untyped third-party libraries; the new test files
  add no errors of their own)
• yamllint — unchanged (no YAML touched)

Notes for reviewers

• The settings-stub cases use
  monkeypatch.setattr(main.settings, "NODE_ROLES", ..., raising=False): the
  keys are not real attributes on the Dynaconf object, so raising=False is
  required, and monkeypatch's teardown restores the absent → default-fallback
  state.
• The make_device / make_interface factories are deliberately minimal —
  they expose only the attributes these helpers read. Later Meta: Unit test coverage for netbox_manager/ #232 tiers should
  extend them in place rather than re-deriving device/interface shapes.

Part of #232.

[ideaship]

No major issues.

[ideaship]

These tests are non-hermetic w.r.t. role settings. Because dynaconf reads the NETBOX_MANAGER_ prefix, a runner with NETBOX_MANAGER_NODE_ROLES/NETBOX_MANAGER_SWITCH_ROLES exported (a legitimate deployment env) makes get_node_roles()/get_switch_roles() return the deployment list, and the is _DEFAULT_* / default-role assertions fail — reproduced as 5 failed, 58 passed. Please isolate the settings here: add os.environ.pop("NETBOX_MANAGER_NODE_ROLES", None) / ..._SWITCH_ROLES alongside the existing setdefaults, or force the state per-test with monkeypatch.delattr(main.settings, "NODE_ROLES", raising=False) in the _absent cases (test_helpers.py:25,48) and the default-role loopback cases (test_loopback_gate.py:54,58,63).

[ideaship]

This test feeds a two-device cable task, but _generate_portchannel_tasks() (netbox_manager/main.py:1870, emit sites 2044/2058/2073/2088) only ever produces single-device device_interface tasks (LAG creations + member assignments), and _filter_portchannel_tasks_by_device is called solely on that list (main.py:2535). The "keeps a pair if either matches" path is therefore unreachable for real inputs — the test pins a fictional shape. Two things to fix: (1) test the real single-device device_interface shapes through this filter; (2) note that _filter_portchannel_tasks_by_device (main.py:2452-2461) is byte-identical to _filter_tasks_by_device (main.py:2430-2439) — fold them into one function in a follow-up, after which this test class collapses into TestFilterTasksByDevice. Also please confirm the intended semantics: is pair-preservation actually required? If so, the current duplicate doesn't provide it and that gap is hidden by this test.

[ideaship]

This is an exact duplicate of test_node_role_always_gets_loopback above — same role_slug="control", same empty custom_fields, same is True assertion — so it adds no coverage. To make it earn its name (node-role eligibility is hwsku-independent), give it a non-empty custom_fields (e.g. {"sonic_parameters": {"hwsku": "AS7326"}}) and assert it's still True. Otherwise drop it.