End-to-end tests on a local model, and in CI on every PR

- Every e2e role runs on qwen3.8-27b on the RTX 4090 (e2e_roles();
  NISSE_E2E_MODEL overrides, e.g. claude-haiku-4-5). The first full run
  passed 7 of 8 in 26 minutes for $0.0000; the eighth was a real finding,
  fixed with the new-project PR.
- scripts/e2e.sh reports a run's spend on the test key and the day's total
  against its budget: above $0 on the local model means a paid fallback.
- CI runs the suite as its own job on pull requests (not pushes to main),
  on a throwaway sshd (--local), cancelling a run a newer push supersedes,
  so the checks job still reports in minutes.
- The plan says so: e2e on every PR; locally only -k / --lf.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

scripts/e2e.sh

@@ -1,13 +1,16 @@
 #!/usr/bin/env bash
 # The end-to-end test (Development Plan M1.6): the whole agent against real
-# services, driven by a scripted operator. Run after each milestone; it
-# spends model tokens and takes minutes, so it is not part of check.sh or CI.
+# services, driven by a scripted operator. CI runs it on every PR (with
+# --local); locally, run only the tests you're writing (-k) or diagnosing
+# (--lf).
 #
 #   scripts/e2e.sh            on the bring-up build VM (configured below)
 #   scripts/e2e.sh --local    on a throwaway local sshd instead
 #
 # Uses tests/integration/nisse-test.yaml and its secrets (.env at the repo
-# root), like the integration tests.
+# root), like the integration tests. Every role runs on a local model
+# (tests/integration/e2e_support.py; NISSE_E2E_MODEL overrides it), so tell
+# the operator before a full run: it keeps the RTX 4090 busy for a while.
 set -euo pipefail
 cd "$(dirname "$0")/.."
 
@@ -19,4 +22,27 @@ else
 fi
 # Each run's state store is kept here, for nisse_dev.dump after a failure.
 export NISSE_E2E_RUNS=${NISSE_E2E_RUNS:-/root/nisse-e2e/test-runs}
-exec uv run pytest -m e2e -rs tests/integration/test_end_to_end.py tests/integration/test_end_to_end_loop.py "$@"
+
+# What the run spent on the test key, read from LiteLLM (spend only). The
+# end-to-end tests run on a local model, so anything above $0 means a call
+# went to a paid one: an override, or a proxy fallback.
+spend() {
+    (set -a; [ -f .env ] && . ./.env; set +a
+     uv run python -c '
+import os, httpx
+r = httpx.get("https://litellm.dongar.net/key/info", timeout=30,
+              headers={"Authorization": "Bearer " + os.environ["NISSE_LLM_TOKEN"]})
+i = r.json()["info"]
+print(i["spend"], i.get("max_budget"))' 2>/dev/null) || echo "? ?"
+}
+read -r before budget < <(spend)
+status=0
+uv run pytest -m e2e -rs tests/integration/test_end_to_end.py tests/integration/test_end_to_end_loop.py "$@" || status=$?
+read -r after _ < <(spend)
+python3 -c "
+b, a, cap = '$before', '$after', '$budget'
+try:
+    print(f'e2e spend: \${float(a) - float(b):.4f} this run; \${float(a):.2f} of \${float(cap):.2f} today')
+except ValueError:
+    print('e2e spend: unknown (could not read the key)')"
+exit "$status"