Fix MARS collector against a real agent's object model

Checked with Get-MarsCollectorDiagnostics.ps1 output from MARS agent
2.0.9274 (VA-FS-01):

- JobStatus.StartTime/EndTime are UTC values not marked as UTC
  (DisplayTime is the local one); they were read as local, putting
  runs 6 hours off. Lookback filtering now compares in UTC too
- ErrorInfo is always present with ErrorCode 0 on success; every
  completed backup would have been a warning. Only non-zero error
  codes count
- Schedule times are TimeSpans; casting them to DateTime failed, so
  next_run_at was always empty
- DatasourceStatus.Datasource is an object; use its DataSourceName
- Also report FileProgress.Failed file counts and FailedFileLog
- Mock rebuilt on the real shapes; diagnostics handle no System State
  policy

Co-Authored-By: Claude Opus 5.5 <[email protected]>

collectors/mars/README.md

@@ -4,10 +4,10 @@ A PowerShell script that runs on each server with the Microsoft Azure Recovery S
 agent and reports its backups to the dashboard. Each server appears as one job,
 named `<SERVER> - Azure Backup` by default.
 
-> **Not yet checked against a real MARS agent.** The cmdlets are the agent's own (`Get-OBJob`,
-> `Get-OBPolicy`, `Get-OBSchedule`), but the property names of the objects they return aren't
-> documented. Before installing on the first server, run `Get-MarsCollectorDiagnostics.ps1` and
-> the collector with `-DryRun` (see below) to confirm what it reads.
+> **Checked against:** MARS agent 2.0.9274, a Files & Folders policy with completed backups only.
+> Not yet seen on a real agent: a failed backup, a backup in progress, or a System State policy.
+> Those paths are written to the same object model, but if one looks wrong, run
+> `Get-MarsCollectorDiagnostics.ps1` while it's in that state and send the output.
 
 ## Install (on the server, as an administrator)
 
@@ -43,9 +43,9 @@ Re-run the installer to change settings or update the script.
 | Company | `Tenant` from the config |
 | Job | One per server, ID `mars:<computer name>`, so changing `JobName` renames rather than duplicates it |
 | Status | Backup job state: Failed → **Failed**. Completed with failed bytes or an error → **Warning**. Aborted/cancelled or unrecognised → **Warning**. In progress → **Running**. Restore jobs are ignored |
-| Started / finished | Job start / end time |
+| Started / finished | Job start / end time. MARS reports these in UTC without marking them as UTC; the collector handles that |
 | Next scheduled | Next slot from the backup policy schedule, or the System State policy's if that's sooner. Empty (so never Overdue) for schedules that run every 2+ weeks, since which week is next can't be told |
-| Message | Job and per-datasource errors, and datasources with failed bytes |
+| Message | Job and per-datasource errors (non-zero error codes), failed file counts, and the failed-file log path |
 | Size | Bytes transferred, summed across datasources |
 
 Each run covers the last 48 hours (`LookbackHours`). Jobs already delivered are recorded in