feat(work): retry a failed item from the shell and the board

[outofcoffee]

Adds spinloop work retry <id> and a retry key on the work board, so a failed item can be put back in the backlog without removing and re-adding it.

Closes #250

Summary

• New work list API path POST /v1/items/{id}/retry: removes a failed item's record, so the run admits it again on a later pass. An item that is not failed gets a 409 naming its state; an unknown id gets a 404.
• New spinloop work retry <id> subcommand, worded and refused the way work abort is.
• The work board retries the selected card with t, and shows t retry in its key hints on a failed card only.
• OpenSpec change add-work-retry updates fleet-orchestrator, work-commands and work-board; docs updated to match.

Implementation details

• Retry reuses what abort already does for a running item: delete the record under the work list's lock and save. No new state shape is stored.
• The items file is not touched, and the failed attempt's kept output stays readable until the new attempt writes over it.
• done items are not retryable; the issue asks for failed ones only.