fix(nvidia): surface Xid 79/154 GPU bus-fall-off as a physical-reboot-required signal

nvidia-bug-report.sh appends .gz to --output-file when gzip is available,
which silently produced empty nvidia-bug-report.txt in support bundles
(cat looked for the uncompressed name that never existed).

Also: a GPU that falls off the PCIe/NVLink bus (Xid 79) or gets flagged
for Node Reboot Required (Xid 154) mid-SAT-run left every downstream test
failing with generic, unrelated-looking errors (CUDA "unknown error",
"unable to determine device handle") with no indication the GPU needed a
physical power-cycle to recover. Detect these codes from SAT run logs and
surface a plain-English "physical reboot required" message in the task's
failure detail, the persisted component-status DB, a dashboard banner on
the Hardware Summary card, and topology diagram GPU-node severity.

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

audit/internal/app/component_status_db.go

@@ -9,6 +9,8 @@ import (
 	"strings"
 	"sync"
 	"time"
+
+	"bee/audit/internal/collector"
 )
 
 // ComponentStatusDB is a persistent, append-only store of hardware component health records.
@@ -235,7 +237,7 @@ func ApplySATResultToDB(db *ComponentStatusDB, target, archivePath string) {
 	dbStatus := satStatusToDBStatus(overall)
 	detail := target + " SAT: " + overall
 	if overall != "OK" {
-		if reason := satFailureDetailFromKV(kv); reason != "" {
+		if reason := prependHardwareFaultBanner(runDir, satFailureDetailFromKV(kv)); reason != "" {
 			detail += " — " + reason
 		}
 	}
@@ -340,7 +342,59 @@ func SATFailureDetail(archivePath string) string {
 	if err != nil {
 		return ""
 	}
-	return satFailureDetailFromKV(parseSATKV(string(data)))
+	reason := satFailureDetailFromKV(parseSATKV(string(data)))
+	return prependHardwareFaultBanner(runDir, reason)
+}
+
+// prependHardwareFaultBanner checks the SAT run directory's captured logs
+// for a known GPU hardware fault (e.g. Xid 79 "fallen off the bus") and, if
+// found, puts a plain-English banner in front of reason — so the task's
+// error message reads "GPU fell off the bus, reboot required" directly
+// instead of just "failed sub-job(s): ...", which tells an engineer nothing
+// without opening the run directory and cross-referencing Xid codes by hand.
+func prependHardwareFaultBanner(runDir, reason string) string {
+	banner := gpuHardwareFaultBanner(runDir)
+	switch {
+	case banner == "":
+		return reason
+	case reason == "":
+		return banner
+	default:
+		return banner + " (" + reason + ")"
+	}
+}
+
+// gpuHardwareFaultBanner scans a SAT run directory's captured *.log files
+// for NVIDIA Xid codes that mean the GPU cannot recover without a physical
+// reboot, returning a de-duplicated, human-readable summary.
+func gpuHardwareFaultBanner(runDir string) string {
+	entries, err := os.ReadDir(runDir)
+	if err != nil {
+		return ""
+	}
+	seen := map[string]bool{}
+	var messages []string
+	for _, e := range entries {
+		if e.IsDir() || !strings.HasSuffix(e.Name(), ".log") {
+			continue
+		}
+		data, err := readFileLimited(filepath.Join(runDir, e.Name()), 2<<20)
+		if err != nil {
+			continue
+		}
+		for _, line := range strings.Split(string(data), "\n") {
+			msg, ok := collector.XidHardwareFaultMessage(line)
+			if !ok || seen[msg] {
+				continue
+			}
+			seen[msg] = true
+			messages = append(messages, msg)
+		}
+	}
+	if len(messages) == 0 {
+		return ""
+	}
+	return strings.Join(messages, "; ")
 }
 
 // satFailureDetailFromKV inspects an already-parsed summary.txt for the

audit/internal/app/support_bundle.go

@@ -227,9 +227,16 @@ fi
 `}},
 	{name: "export/gpu/nvidia-bug-report.txt", cmd: []string{"sh", "-c", `
 if command -v nvidia-bug-report.sh >/dev/null 2>&1; then
-  nvidia-bug-report.sh --output-file /tmp/bee-nvidia-bug-report.log >/dev/null 2>&1 \
-    && cat /tmp/bee-nvidia-bug-report.log \
-    && rm -f /tmp/bee-nvidia-bug-report.log
+  rm -f /tmp/bee-nvidia-bug-report.log /tmp/bee-nvidia-bug-report.log.gz
+  nvidia-bug-report.sh --output-file /tmp/bee-nvidia-bug-report.log >/dev/null 2>&1
+  if [ -f /tmp/bee-nvidia-bug-report.log.gz ]; then
+    gzip -dc /tmp/bee-nvidia-bug-report.log.gz
+  elif [ -f /tmp/bee-nvidia-bug-report.log ]; then
+    cat /tmp/bee-nvidia-bug-report.log
+  else
+    echo "nvidia-bug-report.sh produced no output file"
+  fi
+  rm -f /tmp/bee-nvidia-bug-report.log /tmp/bee-nvidia-bug-report.log.gz
 else
   echo "nvidia-bug-report.sh not found"
 fi