feat(mqtt): store F-070 crash detail and group crashes fleet-wide

Firmware F-070 adds a fuller `crash` object (backtrace, backtrace_corrupted,
elf_sha256) and a new `pre_crash` snapshot (uptime, heap stats, optional
abort_msg) to boot_report and to telemetry.get_boot_history entries.

- Migration c9d0e1f2a3b4: device_boot_events gains crash / pre_crash JSONB,
  stored verbatim so later additive fields need no migration. The legacy
  crash_* columns are still filled; old rows and old firmware are unchanged.
  JSON is bound as CAST(CAST(:x AS TEXT) AS JSONB): a bare JSONB cast makes
  asyncpg JSON-encode the already-encoded string a second time.
- boot_report ingestion passes both objects through.
- telemetry.get_boot_history replies (control/ack) are merged into boot
  history whoever sent the command: an entry matches an existing row on
  boot_count + reset_reason + time within 30 min (boot_count alone is not
  unique - the counter gets reset), and only fills crash/pre_crash the row
  lacks; unmatched entries are boots we never saw live and are inserted at
  the device's timestamp; entries without ts are skipped.
- GET /api/mqtt/crash-groups: fault boots grouped by abort_msg with hex
  addresses stripped, else task + exception cause, else reset reason. When
  abort_msg is present, pc/exc_cause describe abort() itself and are
  ignored for grouping.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

backend/database/__init__.py

@@ -23,6 +23,8 @@ from database.pg_mqtt import (
     insert_boot_event,
     get_boot_events,
     get_latest_boot_event,
+    merge_device_boot_history,
+    get_crash_events,
     insert_ping_sample,
     get_ping_samples,
     get_latest_ping_samples,
@@ -74,6 +76,8 @@ __all__ = [
     "insert_boot_event",
     "get_boot_events",
     "get_latest_boot_event",
+    "merge_device_boot_history",
+    "get_crash_events",
     "insert_ping_sample",
     "get_ping_samples",
     "get_latest_ping_samples",