Add test inventory + checkpoint test runner

docs/testing/test_inventory.md: honest, current list of what automated
coverage actually exists (38 backend pytest tests across 2 files, 0 Godot
tests merged yet) and what doesn't (no integration/endpoint tests at all,
no coverage yet for the in-review RS256/moderation work) -- verified
against a genuinely fresh checkout + fresh Postgres, not just read off the
file tree.

scripts/run_checkpoint_tests.py: runs the whole inventory in one shot --
throwaway Postgres, migrations, pytest, then starts the backend server and
runs every scripts/smoke_test_*.gd against it headlessly. Verified working
both with and without a smoke test present (temporarily tested against
PC1's not-yet-merged scripts/api_client.gd + smoke_test_api.gd, removed
before this commit -- that PR carries its own files).

docs/testing/README.md: what running this at a Package Checkpoint review
does and doesn't prove, and how to grow the inventory as real coverage
gets added.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs/testing/test_inventory.md

@@ -0,0 +1,78 @@
+# Test Inventory
+
+What automated coverage actually exists, what it actually verifies, and what
+doesn't have coverage yet. Keep this current as tests are added — it's the
+thing a Package Checkpoint review (see `docs/testing/README.md`) reads first
+to know what "run the tests" actually checked.
+
+Run everything below in one shot with `scripts/run_checkpoint_tests.py`.
+
+Last verified against `main`: 2026-08-19 (commit `1c75367`).
+
+## Backend — `backend/tests/` (pytest)
+
+Run: `cd backend && python -m pytest -v` (needs `DATABASE_URL` pointed at a
+reachable Postgres — see the runner script; most of these tests don't
+actually touch the DB, but the app's import chain requires a valid URL to
+construct).
+
+**Current total: 38 tests, all passing on a clean `main` checkout.**
+
+| File | Tests | What it covers |
+|---|---|---|
+| `test_auth_age_gate.py` | 10 | `calculate_age()` date-math correctness (exact boundary, leap years, birthday-not-yet-this-year) + `RegisterRequest` date-of-birth validation (parametrized valid/invalid date rejection). Backs the 18+ registration gate (PR #23). |
+| `test_gltf_metadata.py` | 21 | glTF/GLB binary + JSON parsing: magic-byte detection, chunk-padding handling, mesh/animation/texture-reference metadata extraction, malformed-input rejection. Backs the asset-import pipeline (PR #27). |
+
+**What these are, honestly:** pure unit tests of standalone functions
+(`calculate_age`, the glTF parser, Pydantic validators). None of them spin
+up the FastAPI app or hit a real endpoint — they run in ~1s and don't
+actually need Postgres reachable at all (the import chain just needs a
+syntactically valid `DATABASE_URL`, nothing connects during collection).
+
+**Known coverage gaps** (real gaps, not placeholders — nothing enforces filling
+these yet):
+- No integration/endpoint tests at all — `/register`, `/login`, `/refresh`,
+  `/worlds`, `/assets/*` have zero tests that actually call them through
+  the FastAPI app (e.g. via `httpx.AsyncClient` + `ASGITransport`) against a
+  real (or test) database. Everything currently shipped works because it
+  was manually verified at merge time (checkpoint review), not because a
+  test asserts it.
+- No tests for the RS256 auth upgrade (key generation, refresh-token
+  issuance/persistence) — PR #30, pending review as of this writing,
+  manually verified (correct key usage, all 38 existing tests still pass)
+  but not covered by an automated test of its own yet.
+- No tests for the moderation schema/migration (in progress as of this
+  writing) beyond "the migration runs."
+
+## Godot client — `scripts/smoke_test_*.gd`
+
+**Nothing merged into `main` yet as of this writing** — `sandbox_probe/test_script.gd`
+is the only Godot script on `main` right now, and it's **not a real
+test** (a leftover manual probe, returns a hardcoded string, asserts
+nothing). Don't count it toward coverage; delete or repurpose it if anyone
+actually needs a sandbox probe again.
+
+`scripts/smoke_test_api.gd` (PC1, PR #29, pending review as of this writing)
+will be the first real one once merged — `ApiClient` autoload's
+`health_check()` against a **live backend** (set `AGP_API_BASE_URL` to point
+at one), verified by hand end-to-end before opening the PR: exercises the
+real HTTP request path (`HTTPRequest` node lifecycle, JSON parse, the
+result-dictionary contract), not just that the script compiles. Exits 1 on
+failure, so it's CI/script-friendly.
+
+Run (once merged): `godot --headless --path . --import` once (builds the
+class cache — skip this and every `class_name`-declared class falsely
+errors as undeclared), then for each `scripts/smoke_test_*.gd`:
+`godot --headless --path . --script res://scripts/<name>.gd`
+
+**Known gap once it lands:** only `health_check()` is exercised end-to-end.
+`login`, `register`, `get_worlds`, `get_me` are implemented (see
+`scripts/api_client.gd`) but have no smoke coverage yet — each would need a
+real user/world fixture in the target backend to test against meaningfully.
+
+## What "full coverage" should eventually mean here
+
+Not defined precisely yet — the honest target as of 2026-08-19 is just
+"more than pure-function unit tests." As real integration/endpoint tests get
+added, update the tables above rather than letting this doc drift stale;
+a wrong inventory is worse than no inventory.