test: pin query-count baselines for tag inheritance hot paths (#14811)

* test: pin query-count baselines for tag inheritance hot paths

Adds unittests/test_tag_inheritance_perf.py with assertNumQueries baselines
on the six hottest tag inheritance paths (Product tag add/remove propagating
to N findings, child create under inheritance, sticky enforcement on child
tag edits). Numbers are pinned against current `dev` behavior so subsequent
optimization work shows up as concrete query-count reductions instead of
relying on manual benchmarking.

The class is intentionally temporary: pins move down as the redesign work
lands and the file can be deleted once the targets are met.

* test: add Endpoint (V2) and Location (V3) propagation baselines

Extends the perf test class with two more pinned hot paths so all child
models exercised by `propagate_tags_on_product_sync` are covered:

  product_tag_add  -> 100 endpoints (V2) : 3958
  product_tag_remove -> 100 endpoints (V2): 3740
  product_tag_add  -> 100 locations (V3) : 4532
  product_tag_remove -> 100 locations (V3): 4307

Both V2 and V3 paths run regardless of the ambient `V3_FEATURE_LOCATIONS`
setting via per-test `@override_settings(...)`. CI matrix runs the suite
in both modes, so dynamic pin selection (`_pin(v2=..., v3=...)`) handles
the small per-mode count differences on the existing finding tests.

* test: add ZAP import/reimport baselines + V2/V3 variants for every scenario

Two additions:

1. New TagInheritanceImportPerfBaselines class pins query counts for the
   importer hot path (production's heaviest tag-inheritance scenario).
   Both first-import and no-change-reimport are covered, each with V2
   and V3 method variants:

     zap_import_v2                : 1461
     zap_import_v3                : 1319
     zap_reimport_no_change_v2    : 77
     zap_reimport_no_change_v3    : 95

2. Restructures the existing baseline class so every scenario has both
   a _v2 and _v3 method variant via per-test @override_settings. The
   whole suite now runs both modes in a single invocation; no need to
   run twice with different DD_V3_FEATURE_LOCATIONS env.

Phase A leaves the importer numbers ~unchanged (importer hot loop is
creation-driven, not the bulk-propagation path Phase A targets). Phase
B's tag_inheritance.batch() context manager is the lever that lowers
these numbers.

* test: warm ContentType cache in tag inheritance perf baselines

First V3 Location op in the class paid a one-time ContentType lookup,
producing a matrix-dependent off-by-one (V3-default-on CI: 4531;
V3-default-off CI + local: 4532). Match the warm-up pattern used in
test_importers_performance and pin EXPECTED_PRODUCT_TAG_ADD_100_LOCATIONS
to the post-warm value (4531).

Author: valentijnscholten