feat(dashboard): mechanically link spawned/torn-down tasks to their card

[joliverMI]

Summary

The fleet auditor traced every status discrepancy it has raised to one cause: no card's agent/backlog_ref was ever populated, so a status transition only happened when someone remembered to make it by hand - worst of all at teardown, the exact moment the worker doing the remembering is removed.

• bin/fm-spawn.sh --card <card-id> (new flag): writes dashboard_card=<card-id> into the task's own state/<id>.meta (the durable identity link teardown later consumes), then best-effort sets the card's backlog_ref to <home>:<task-id> (the exact format the fleet auditor's sweep already reads) and its agent to the task id, advancing a not_started card to working. Refused on --secondmate and --relaunch for the same reason --mode/--yolo are: those spawns don't own that axis.
• bin/fm-teardown.sh: reads dashboard_card= back out of meta and, once a real (non---force) landing actually succeeds, advances the card to testing - the exact hand-off the audit named as most likely to be forgotten. Scout and secondmate teardowns never touch the card (scratch worktree / not a backlog item), a --force discard never touches it (a discard is not a landing), and an already-complete card is never downgraded back.
• Reuses the board's existing ref/agent/status fields end to end - no new store.
• No card is the normal case and stays a complete no-op - the vast majority of fleet work never touches the dashboard. A bad card id or an unreachable dashboard never fails the spawn or teardown; it prints a loud warning: and best-effort records fm-dashboard.sh audit-log --fleet, so a link that could not fire is visible in the same discrepancy log the fleet auditor already reads, not silently dropped.
• docs/dashboard.md gets a new "The mechanical card link" section owning the full design/rationale; the fleet-dashboard skill gets a one-paragraph pointer to prefer --card over manual follow-up calls.

Backfill

All 21 live cards still carry no agent/backlog_ref. None were backfilled - here's why, rather than guessing:

• 20 cards (captain_dj, e.g. hue-area-selection-c10f, three-mode-display-port-8lvy, ...) belong to a project this task has no access to. Their own notes/prompts carry no task-id evidence, and I have no filesystem access to that project's backlog/state to independently establish which task, if any, currently serves each card.
• 1 card (fleet-dashboard-gh63, captain firstmate) is already complete with no live task - the work that shipped it has presumably already been torn down, so there is nothing left to link retroactively.

Backfilling is now mechanical going forward: pass --card <id> the next time a task is dispatched to serve one of these.

Test plan

• bin/fm-lint.sh bin/fm-spawn.sh bin/fm-teardown.sh tests/fm-dashboard-card-link.test.sh - shellcheck clean
• New tests/fm-dashboard-card-link.test.sh (9 cases, real dashboard server + real fm-spawn.sh/fm-teardown.sh invocations, no mocked dashboard): link+advance-to-working at spawn, no-card no-op, unreachable-dashboard doesn't fail spawn/teardown, unknown-card-id warns + records a fleet finding, advance-to-testing at teardown on a real landed local-only merge, force-discard never advances, already-complete is never downgraded
• Full existing suites re-run clean with zero regressions: tests/fm-dashboard.test.sh, tests/fm-teardown.test.sh (62/62), tests/fm-spawn-worktree-settle.test.sh, tests/fm-spawn-batch.test.sh, tests/fm-spawn-dispatch-profile.test.sh, tests/fm-teardown-endpoint-safety.test.sh, tests/fm-trace-context-spawn.test.sh
• bin/fm-doc-audience-check.sh clean after the docs/dashboard.md addition