gpu: collect reset-required/row-remap status and classify Xid codes by severity

nvidia-smi exposes reset_status.reset_required and remapped_rows.* only on
newer drivers for Ampere+ GPUs; queried via a separate exec call since an
unrecognized field name fails the whole --query-gpu command and would have
wiped out unrelated telemetry (temp/ECC/power) on older drivers otherwise.

Also refines Xid severity in both the ingest dmesg collector and the
always-on kmsg watcher: Xid 64 (row-remap InfoROM write failure) now
escalates to Critical instead of the generic warning every other Xid got,
and fixes the SAT-window flush path which previously hardcoded "Warning"
and ignored pattern severity entirely.

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

audit/internal/collector/nvidia_test.go

@@ -124,5 +124,74 @@ func TestEnrichPCIeWithNVIDIAData_driverMissingFallback(t *testing.T) {
 	}
 }
 
+func TestParseNVIDIAReliability(t *testing.T) {
+	raw := "0000:65:00.0, No, 0, 2, No, Yes\n"
+	byBDF, err := parseNVIDIAReliabilityCSV(raw)
+	if err != nil {
+		t.Fatalf("parse failed: %v", err)
+	}
+
+	gpu, ok := byBDF["0000:65:00.0"]
+	if !ok {
+		t.Fatalf("gpu by normalized bdf not found")
+	}
+	if gpu.ResetRequired == nil || *gpu.ResetRequired {
+		t.Fatalf("reset_required: got %v, want false", gpu.ResetRequired)
+	}
+	if gpu.RemapCorrectable == nil || *gpu.RemapCorrectable != 0 {
+		t.Fatalf("remap correctable: got %v", gpu.RemapCorrectable)
+	}
+	if gpu.RemapUncorrectable == nil || *gpu.RemapUncorrectable != 2 {
+		t.Fatalf("remap uncorrectable: got %v", gpu.RemapUncorrectable)
+	}
+	if gpu.RemapPending == nil || *gpu.RemapPending {
+		t.Fatalf("remap pending: got %v, want false", gpu.RemapPending)
+	}
+	if gpu.RemapFailure == nil || !*gpu.RemapFailure {
+		t.Fatalf("remap failure: got %v, want true", gpu.RemapFailure)
+	}
+}
+
+func TestEnrichPCIeWithNVIDIAData_resetRequiredCritical(t *testing.T) {
+	vendorID := NvidiaVendorID
+	bdf := "0000:65:00.0"
+	devices := []schema.HardwarePCIeDevice{
+		{VendorID: &vendorID, BDF: &bdf},
+	}
+
+	byBDF := map[string]nvidiaGPUInfo{
+		"0000:65:00.0": {ResetRequired: ptrBool(true)},
+	}
+
+	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
+	if out[0].Status == nil || *out[0].Status != statusCritical {
+		t.Fatalf("status: got %v, want %v", out[0].Status, statusCritical)
+	}
+	if out[0].ResetRequired == nil || !*out[0].ResetRequired {
+		t.Fatalf("reset_required: got %v", out[0].ResetRequired)
+	}
+}
+
+func TestEnrichPCIeWithNVIDIAData_remapFailureCritical(t *testing.T) {
+	vendorID := NvidiaVendorID
+	bdf := "0000:65:00.0"
+	devices := []schema.HardwarePCIeDevice{
+		{VendorID: &vendorID, BDF: &bdf},
+	}
+
+	byBDF := map[string]nvidiaGPUInfo{
+		"0000:65:00.0": {RemapFailure: ptrBool(true)},
+	}
+
+	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
+	if out[0].Status == nil || *out[0].Status != statusCritical {
+		t.Fatalf("status: got %v, want %v", out[0].Status, statusCritical)
+	}
+	if out[0].RemappedRowsFailure == nil || !*out[0].RemappedRowsFailure {
+		t.Fatalf("remapped_rows_failure: got %v", out[0].RemappedRowsFailure)
+	}
+}
+
 func ptrInt64(v int64) *int64     { return &v }
 func ptrFloat(v float64) *float64 { return &v }
+func ptrBool(v bool) *bool        { return &v }