[Fix] Platform alerts are missing when sandbox capacity blocks task starts

[roomote-roomote]

Opened on behalf of Matt Rubens. Follow up by mentioning @roomote-roomote, in the web UI, or in Slack.

What changed

• Detect Modal sandbox-creation failures caused specifically by the concurrent sandbox limit.
• Persist and deliver a manager-facing platform issue alert even though the worker never started.
• Keep alert delivery best-effort so a Slack delivery problem cannot prevent the task run from being finalized as failed.

Why this change was made

Platform issue alerts previously depended on a running agent invoking the reporting tool. Concurrent sandbox limits prevent the worker from starting, so these operational failures could not reach the configured Slack alert destination.

Impact

Operators now receive the configured platform issue alert when Modal capacity blocks a task start. Other RESOURCE_EXHAUSTED errors remain excluded, and the originating task still follows the normal failed-start lifecycle.

[roomote-roomote]

No code issues found. See task

_{Reviewed c482ced}