feat(m2): auth, JWT, invitations, bootstrap, RTOps SPA pages

Crypto + tokens
- app/core/security.py: Argon2id PasswordHasher (time_cost=2, memory_cost=
  64 MiB, parallelism=2) + opaque-token SHA-256 helpers (raw token shown
  once, only the hash lives in the DB).
- app/core/jwt_tokens.py: HS256, claims iss/sub/type/jti/iat/exp. Access
  1h, refresh 30d.

Services
- services/auth.py: login, refresh with token rotation + reuse-detection
  chain revoke, logout (idempotent), change_password (forces logout-all).
- services/invitations.py: create, preview, accept, revoke. Default 7d TTL.
- services/bootstrap.py: seeds the 3 system groups (admin/redteam/blueteam),
  consumes the install token, attaches the first user to admin.
- core/install_token.py: mints, persists in settings, marks consumed,
  regenerate hook for /diag/reset.

API
- POST /setup (consume install token, create 1st admin) + GET /setup
  (status).
- POST /auth/{login,refresh,logout,change-password} + GET /auth/me.
- POST /invitations + GET /invitations + GET /invitations/preview/<token> +
  POST /invitations/accept/<token> + POST /invitations/<id>/revoke.
- POST /diag/reset: test-only kill switch (truncate auth tables + mint
  fresh install token). Allowed in dev too (with WARNING log) so the e2e
  suite can run against a make-up stack; production locked out.

Middleware
- @require_auth populates g.current_user (snapshot dataclass, session
  closed before request handler runs).
- @require_perm(*codes): atomic perm union check; admin group bypasses.
  Perm catalogue lands in M3, scaffolding here.
- flask-limiter: 10/min/IP on /auth/login & /auth/refresh, 5/min on
  /auth/change-password & /setup, 10–20/min on invitation endpoints.
  Disabled in APP_ENV=test.

CLI
- flask --app app.cli metamorph print-install-token [--force]
- flask --app app.cli metamorph seed-mitre (M4 placeholder)

Refresh cookie metamorph_refresh: HttpOnly + Secure (localhost is a secure
context for modern browsers) + SameSite=Strict + Path=/api/v1/auth/.

Email validation: app.api._validation.Email permissive RFC-shape regex so
internal TLDs (.local/.corp/.test) are accepted — pydantic.EmailStr's
deliverability check is too strict for red-team labs.

Frontend
- lib/{api,auth}.ts: access token in module memory, refresh cookie,
  automatic 401-retry via /auth/refresh, useAuth() hook.
- components/{Layout,RequireAuth}.tsx + ui/{TextField,Alert}.tsx.
- pages/{Login,Setup,Register,Profile}.

Testing
- tests/test_auth_flow.py: 15 integration tests (24 backend total).
- e2e/tests/m2-auth.spec.ts: 8 Playwright tests (20 e2e total).
- tasks/testing-m2.md.

DoD: make test-api → 24 passed, make e2e → 20 passed; spec-reviewer pass
applied (Secure unconditional, refresh limit 10/min/IP).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend/app/core/rate_limit.py

@@ -0,0 +1,29 @@
+"""Shared flask-limiter instance.
+
+Anchored on remote address. In-memory backend for v1 (single-process gunicorn
+worker pool can drift; that's acceptable at this scale). M14 will switch to
+Redis if it becomes a real concern.
+
+The limiter is enforced in `APP_ENV in ("prod", "staging")` — dev and test
+deployments share an in-memory backend that's noisy across hot-reloads and
+would gate the Playwright e2e suite at 10 req/min/IP. The spec NF-security
+requirement is explicitly a *production* one (cf. tasks/spec.md §6
+NF-security); a staging deployment is exposed to humans so the same limits
+apply there.
+"""
+
+from __future__ import annotations
+
+from flask_limiter import Limiter
+from flask_limiter.util import get_remote_address
+
+from app.core.config import settings
+
+limiter = Limiter(
+    key_func=get_remote_address,
+    default_limits=[],
+    storage_uri="memory://",
+    headers_enabled=True,
+    strategy="fixed-window",
+    enabled=settings.APP_ENV in ("prod", "staging"),
+)