fix: stops orphan tagging redfish inventory items the source still manages

[marcinpsk]

Problem

tag_all_the_things() adds the orphaned tag to every object which carries the primary tag and which a run does not register with its source. Two paths in the check_redfish source left inventory items untouched, so they were tagged orphaned while the source was still managing them.

1. An empty batch said nothing. update_all_items() read the component type back from the first item in the batch and returned early on an empty list:

if len(items) == 0:
    return

inventory_type = grab(items, "0.inventory_type")

A scan which reports no component of a type therefore left every one of that device's components of that type untouched, instead of marking them absent. A server whose fans stop being reported is the case which shows it.

2. Already absent items were never touched again. Marking an item absent was conditional on its health actually changing:

elif grab(nb_inventory_item, "data.custom_fields.health") != "Absent":

So an item already at absent was skipped on every later run, and tagged orphaned from then on.

Fix

The component type is passed in by the caller, which knows it whether or not the batch is empty, and is stamped onto the items in update_all_items(), so the value used for the lookup and the value stored on each item cannot drift apart.

Registering the item with the source is now unconditional. update() still writes nothing when the value is unchanged, so this adds no NetBox requests.

Notes

• Changes the update_all_items() signature and drops the per-item inventory_type key at the nine call sites.
• Adds [tool.pytest.ini_options] to pyproject.toml
• The make_source() helper - should be moved to tests/conftest.py when available (I see it in adds a vcsim based test suite and runs it in CI #541)