[adamtheturtle]

Shared target-summary and recognition-report tests currently pass a concrete backend identity to a seeding helper that imports backend setup. Supply seeding through a typed seed_recognition_counts pytest fixture instead, so the call sites only request the operation they need. Backend setup supplies either the running in-memory mock's operation or the target-manager HTTP request.

This four-file change follows #3742 and advances #3738, part of #3737. Further backend decoupling remains in #3738.

Validation:

• Python 3.15: all 60 recognition-report and seeded-summary tests pass across the in-memory and Flask backends, using synthetic credentials with retries disabled.
• The new fixture's executable lines and both backend branches are covered. The operation protocol and both changed test modules have 100% coverage in the targeted run.
• Pre-commit and pre-push hooks pass, including CI partition validation, Mypy, Pyright, public type completeness, Ty, and Pyrefly. Pylint passes for all changed files, and git diff --check passes.

The local run is targeted; full-suite coverage and real-Vuforia validation remain for hosted CI.