fix(parser): support Inspur HGX dump_<serial>_<timestamp>/ onekeylog layout

Fixes #20. This onekeylog variant has no devicefrusdr.log at all: FRU/sensors
come from raw ipmitool text output, PCIe/GPU presence has a dedicated
structural snapshot, SEL lives at a different path, and BMC component
failures are logged separately from SEL/IDL.

- Fall back to component/fru.txt (same FRU block format as devicefrusdr.log)
  and component/sensor.txt / sdr.txt (ipmitool sensor list / sdr elist)
  when devicefrusdr.log is absent.
- Parse log/bmc/diagnose/OtrdDiagnoseComponent.json's PCIe Device Info array
  for GPU/PCIe presence and link state, independent of SEL/IDL alarm
  history; flag devices running below their negotiated max link
  speed/width as degraded with a Warning event.
- Fall back to log/sel.csv (same format as selelist.csv) when selelist.csv
  is absent.
- Parse log/bmc/commer-comp/{commerslot,commerhmc,commerswvr,commerswcpld}
  logs (including rotated *.tar.gz.N parts) into failure events, filtering
  known-noisy lines.
- Collapse SEL events duplicated across sources by
  (timestamp, event_type, description).
- Surface a CollectionError when FRU/sensors are still empty after all
  fallbacks, instead of silently returning an empty inventory.
- Fix ParseFRU: a later placeholder "Product Serial : 0" / "Product Part
  Number : NULL" line in the same FRU block (e.g. SCM_FRU) was overwriting
  an already-parsed real Board Serial/Part Number.

Verified against dump_23DB01633_20260727-1359.tar.gz (HGX B200, KR9288-X3):
fru 0→21, sensors 0→303, 8 GPUs present at Gen5 x16 in slots 100-107.

Deferred (not covered by this change): BIOS-change-settings context and
BIOS POST codes from the same layout — see bible-local/10-decisions.md
ADL-048.

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

internal/parser/vendors/inspur/diagnose.go

@@ -0,0 +1,121 @@
+// Parses log/bmc/diagnose/OtrdDiagnoseComponent.json, an HGX BMC structural
+// snapshot of PCIe topology (present GPUs/NICs/NVMe, negotiated link state).
+// This is a direct point-in-time source for GPU presence and PCIe link
+// degradation, independent of SEL/IDL alarm history.
+package inspur
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"git.mchus.pro/mchus/logpile/internal/models"
+	"git.mchus.pro/mchus/logpile/internal/parser/vendors/pciids"
+)
+
+// otrdDiagnosePCIeEntry mirrors the relevant fields of "Pcie Device Info"
+// entries in OtrdDiagnoseComponent.json.
+type otrdDiagnosePCIeEntry struct {
+	LocString        string `json:"LocString"`
+	PcieSlot         int    `json:"PcieSlot"`
+	PresentStatus    int    `json:"PresentStatus"`
+	VendorId         int    `json:"VendorId"`
+	DeviceId         int    `json:"DeviceId"`
+	BusNumber        int    `json:"BusNumber"`
+	DeviceNumber     int    `json:"DeviceNumber"`
+	FunctionNumber   int    `json:"FunctionNumber"`
+	CurrentLinkSpeed int    `json:"CurrentLinkSpeed"`
+	MaxLinkSpeed     int    `json:"MaxLinkSpeed"`
+	NegotiatedLinkWidth int `json:"NegotiatedLinkWidth"`
+	MaxLinkWidth     int    `json:"MaxLinkWidth"`
+	SerialNumber     *string `json:"SerialNumber"`
+	PartNumber       *string `json:"PartNumber"`
+}
+
+type otrdDiagnoseComponent struct {
+	PcieDeviceInfo []otrdDiagnosePCIeEntry `json:"Pcie Device Info"`
+}
+
+// ParseOtrdDiagnosePCIe parses the "Pcie Device Info" array from
+// OtrdDiagnoseComponent.json into PCIe device inventory entries.
+func ParseOtrdDiagnosePCIe(content []byte) []models.PCIeDevice {
+	var doc otrdDiagnoseComponent
+	if err := json.Unmarshal(content, &doc); err != nil {
+		return nil
+	}
+
+	devices := make([]models.PCIeDevice, 0, len(doc.PcieDeviceInfo))
+	for _, e := range doc.PcieDeviceInfo {
+		present := e.PresentStatus == 1
+		_, deviceName := pciids.DeviceInfo(e.VendorId, e.DeviceId)
+		deviceClass := normalizeModelLabel(deviceName)
+		if strings.Contains(strings.ToUpper(deviceName), "NVIDIA") {
+			deviceClass = "GPU (" + deviceName + ")"
+		}
+
+		serial := ""
+		if e.SerialNumber != nil {
+			serial = strings.TrimSpace(*e.SerialNumber)
+		}
+		partNum := ""
+		if e.PartNumber != nil {
+			partNum = strings.TrimSpace(*e.PartNumber)
+		}
+
+		status := ""
+		if present && isLinkDegraded(e) {
+			status = "Link Degraded"
+		}
+
+		devices = append(devices, models.PCIeDevice{
+			Slot:         e.LocString,
+			VendorID:     e.VendorId,
+			DeviceID:     e.DeviceId,
+			BDF:          formatBDF(e.BusNumber, e.DeviceNumber, e.FunctionNumber),
+			DeviceClass:  deviceClass,
+			Manufacturer: normalizeModelLabel(pciids.VendorName(e.VendorId)),
+			LinkWidth:    e.NegotiatedLinkWidth,
+			LinkSpeed:    fmt.Sprintf("GEN%d", e.CurrentLinkSpeed),
+			MaxLinkWidth: e.MaxLinkWidth,
+			MaxLinkSpeed: fmt.Sprintf("GEN%d", e.MaxLinkSpeed),
+			PartNumber:   partNum,
+			SerialNumber: serial,
+			Present:      &present,
+			Status:       status,
+		})
+	}
+
+	return devices
+}
+
+func isLinkDegraded(e otrdDiagnosePCIeEntry) bool {
+	if e.MaxLinkSpeed > 0 && e.CurrentLinkSpeed < e.MaxLinkSpeed {
+		return true
+	}
+	if e.MaxLinkWidth > 0 && e.NegotiatedLinkWidth < e.MaxLinkWidth {
+		return true
+	}
+	return false
+}
+
+// BuildPCIeLinkDegradationEvents emits a Warning event for every PCIe device
+// present in OtrdDiagnoseComponent.json but running below its negotiated max
+// link speed/width (e.g. a GPU baseboard degrading from Gen5 x16 to a lower
+// state), so the log viewer surfaces the same signal the IDL alarms would.
+func BuildPCIeLinkDegradationEvents(devices []models.PCIeDevice) []models.Event {
+	var events []models.Event
+	for _, d := range devices {
+		if d.Present == nil || !*d.Present || d.Status != "Link Degraded" {
+			continue
+		}
+		events = append(events, models.Event{
+			ID:          fmt.Sprintf("pcie_link_degraded_%s", strings.TrimPrefix(d.Slot, "#")),
+			Source:      "PCIe Diagnose",
+			SensorType:  "pcie_link",
+			EventType:   "Link Degraded",
+			Severity:    models.SeverityWarning,
+			Description: fmt.Sprintf("%s: link degraded to %s x%d (max %s x%d)", d.Slot, d.LinkSpeed, d.LinkWidth, d.MaxLinkSpeed, d.MaxLinkWidth),
+		})
+	}
+	return events
+}