feat(health): GPU stress monitor + per-drive SMART health/wear

Two diagnostics for the load-correlated GPU crashes and for storage wear.

GPU stress (`rigdoctor stress` + a System Health "Stress test…" dialog): drive a GPU
load and sample sensors at high rate, then report per-metric min/avg/peak, time spent
above each temp threshold, power vs limit, throttling (decoded from the NVML
clocks-event bitmask), and any GPU fault (Xid / VA-space freeze / query-timeout hang)
in the window. Load source: explicit --command, an auto-detected loader, or
monitor-only (you launch the game). Analysis is a pure, unit-tested function.

Drive health (core/drives.py): parse full `smartctl --json` per drive into prioritized
findings — SMART verdict, derived life-left % (NVMe percentage_used or SATA
wear-leveling), power-on hours, TBW, temperature, and failure predictors
(reallocated/pending/offline sectors, NVMe media errors, low spare). Replaces the old
pass/fail-only check_smart; runs through the same elevated path (collect-priv / sudo),
degrading to "needs root" notes unprivileged.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_stress.py

@@ -0,0 +1,77 @@
+"""Tests for the GPU stress + thermal-monitor analysis (synthetic ticks, no real GPU)."""
+
+import unittest
+
+from rigdoctor.core import stress
+from rigdoctor.core.health import CRITICAL, OK, WARNING
+
+
+def _tick(temp=None, power=None, throttle=(), capped=False, lost=False, dt=1.0, **extra):
+    values = {}
+    if temp is not None:
+        values["gpu.temp"] = temp
+    if power is not None:
+        values["gpu.power"] = power
+    values.update(extra)
+    return stress._Tick(dt=dt, values=values, throttle=list(throttle), power_capped=capped, lost=lost)
+
+
+class SummarizeTests(unittest.TestCase):
+    def test_stable_run_is_ok(self):
+        ticks = [_tick(temp=t, power=200, **{"gpu.power_limit": 280}) for t in (60, 65, 70, 72)]
+        r = stress.summarize(ticks, load="monitor-only", interval=1.0, faults=[])
+        self.assertEqual(r.severity, OK)
+        self.assertEqual(r.peak_temp, 72)
+        self.assertEqual(r.max_power, 200)
+        self.assertEqual(r.power_limit, 280)
+        self.assertFalse(r.throttled)
+        self.assertIn("Stable", r.verdict)
+
+    def test_dwell_time_above_thresholds(self):
+        # 3 ticks of 2s each at 82/86/92 °C → ≥80 for all 6s, ≥85 for 4s, ≥90 for 2s.
+        ticks = [_tick(temp=82, dt=2.0), _tick(temp=86, dt=2.0), _tick(temp=92, dt=2.0)]
+        r = stress.summarize(ticks, load="x", interval=2.0, faults=[])
+        self.assertEqual(r.time_above[80], 6.0)
+        self.assertEqual(r.time_above[85], 4.0)
+        self.assertEqual(r.time_above[90], 2.0)
+        self.assertNotIn(95, r.time_above)  # never reached → omitted
+
+    def test_throttling_is_a_warning(self):
+        ticks = [_tick(temp=88, throttle=["HW thermal slowdown"])]
+        r = stress.summarize(ticks, load="x", interval=1.0, faults=[])
+        self.assertEqual(r.severity, WARNING)
+        self.assertTrue(r.throttled)
+        self.assertIn("HW thermal slowdown", r.throttle_reasons)
+
+    def test_high_temp_without_throttle_is_a_warning(self):
+        r = stress.summarize([_tick(temp=93)], load="x", interval=1.0, faults=[])
+        self.assertEqual(r.severity, WARNING)
+        self.assertIn("hot", r.verdict.lower())
+
+    def test_gpu_lost_is_critical(self):
+        ticks = [_tick(temp=70), _tick(lost=True)]
+        r = stress.summarize(ticks, load="x", interval=1.0, faults=[])
+        self.assertEqual(r.severity, CRITICAL)
+        self.assertTrue(r.gpu_lost)
+
+    def test_journal_fault_is_critical(self):
+        r = stress.summarize([_tick(temp=70)], load="x", interval=1.0,
+                             faults=["NVIDIA Xid 79 ×1"])
+        self.assertEqual(r.severity, CRITICAL)
+        self.assertIn("Xid 79", r.verdict)
+
+    def test_no_telemetry_is_info(self):
+        r = stress.summarize([_tick()], load="monitor-only", interval=1.0, faults=[])
+        self.assertEqual(r.severity, "info")
+        self.assertIsNone(r.peak_temp)
+
+
+class ThrottleDecodeTests(unittest.TestCase):
+    def test_throttle_bits_map_to_reasons(self):
+        # the constants used by _throttle_state decode the NVML active-reasons bitmask
+        self.assertIn("HW thermal slowdown", stress._THROTTLE_BITS.values())
+        self.assertIn("SW thermal slowdown", stress._THROTTLE_BITS.values())
+
+
+if __name__ == "__main__":
+    unittest.main()