Add sandbox startup failure observability and improve error messages (REMOTE-2210)

[seemeroland]

Summary

Addresses REMOTE-2210: Cloud agent sandbox fails to start before startup deadline.

Changes

1. New metric: oz_worker_task_sandbox_startup_failures_total{reason}

Tracks task failures that occur within the first 5 minutes of task assignment, labeled by failure reason. This counter is a subset of oz_worker_task_failures_total designed to isolate systemic sandbox provisioning issues (slow image pulls, cluster resource pressure, admission failures) from long-running task errors. The metric is primed at startup so dashboards and alerts can query it immediately.

Recommended PromQL alert:

sum by (reason) (rate(oz_worker_task_sandbox_startup_failures_total[5m])) > 0

2. Startup window failure detection in executeTask

When a task fails within the sandboxStartupWindow (5 min), the worker now:

• Emits a Warn-level log with task ID, elapsed duration, failure phase, and reason
• Records a task.sandbox_startup_failure trace event with those same attributes
• Increments the new oz_worker_task_sandbox_startup_failures_total counter

3. Improved user-facing error messages

userFacingTaskError now produces actionable, reason-specific messages for known backendFailureError types instead of always falling back to the raw "Failed to execute task: ..." string. Covered reasons:

Reason	New message
active_deadline	"The agent sandbox did not complete before the job deadline was exceeded. This may indicate slow container image pulls, cluster resource pressure, or sandbox startup delays — please try again."
container_create / container_start	"The agent sandbox failed to start. This may indicate a container image issue, insufficient cluster resources, or a configuration problem — please try again."
sidecar_prep	"The agent sandbox failed to prepare its required dependencies..."
image_pull	"The agent sandbox could not pull its container image..."
unschedulable	"The agent sandbox could not be scheduled due to insufficient cluster resources..."
container_oom	"The agent sandbox ran out of memory and was terminated..."

4. README documentation

Added the new metric to the catalog and sample dashboards/alerts section.

---

Conversation: https://staging.warp.dev/conversation/3785204b-d602-4fb7-8a4f-e8a0d934d721
Run: https://oz.staging.warp.dev/runs/019f71e6-7b6c-7eed-95ef-49206074f37f

This PR was generated with Oz.