test(backend): cover the infrastructure layer (db/ + core + main)

[untraceablez]

Why

Part 2 of closing the remaining coverage gaps (after #64 took App.tsx 0%→100%). This covers the backend infrastructure layer — the db/, core/, and main.py modules that the earlier Track B service/router rounds skipped. Test-only; no source changes.

Coverage

File	Before	After
db/base.py	63%	100%
db/seed.py	0%	**100%**¹
core/cache.py	32%	100%
core/deps.py	69%	100%
main.py	74%	100%
db/session.py	47%	97%²

¹ bar the if __name__ == "__main__" entrypoint guard
² the lone remaining line is the non-test pool_size branch, which only runs when ENVIRONMENT != "test"

Notable details

• db/base.py — TimestampMixin.__init__ is never reached through the real models (they all declare (Base, TimestampMixin), so Base.__init__ wins the MRO). An isolated probe model on its own declarative base puts the mixin first to exercise both the default and explicit-value branches — without touching the app's metadata.
• core/cache.py — every CacheService method incl. error branches and the delete_pattern scan_iter path, plus the get_redis singleton and close_redis.
• core/deps.py — all auth-dependency branches: user id/role token parsing, current-user lookup, API-client auth, and require_client_scope. (create_access_token forces type="access", so client-token cases stub decode_token.)
• main.py — proxy-headers middleware (CF-Connecting-IP + X-Forwarded-* paths), the validation + Pantrie exception handlers, and the lifespan custom-domain and db-error branches (via a stubbed get_db, so it doesn't depend on DATABASE_URL vs TEST_DATABASE_URL).

Result

• Backend suite: 399 passed (pantrie-backend container)
• Overall project coverage: 97.8% → 99%

🤖 Generated with Claude Code