Refill independent swarm workers as capacity becomes available

[randomparity]

Problem

Independent fleet jobs can leave capacity idle because the next wave waits for every
worker in the current wave, even when one slow worker is the only remaining task.

Evidence

• skills/summon-swarm/SKILL.md:56 caps concurrency and requires full-wave completion.
• skills/summon-swarm/SKILL.md:74 persists per-worker sessions and status.
• skills/summon-swarm/SKILL.md:82 requires output review; :106 owns failure/recovery.

Expected

After a worker's observed end, review/reconcile its result and refill available capacity
with an already-authorized independent task at the next completion-handling opportunity.
Use the lowest applicable task/workflow/harness concurrency cap, including a cap of one.
Preserve isolated workspaces, output review, fix-round budgets, and replacement rules.
Pending repairs count against capacity; uncertain ownership is reconciled before reuse.
Resume accounts for existing attempts instead of launching duplicate workers.

Proposed approach

Replace the whole-wave dispatch/collection procedure with a bounded ready queue expressed
through the existing fleet manifest and shared monitoring procedure. Keep the skill scoped
to independent volume work; no DAG engine or shared-workspace execution is introduced.
Verify staggered completion with more tasks than slots, one-slot serial execution, failure
and repair, stale reports, and resume. Run a bounded live refill case and preserve its trace
for the saga's final comparison. Follow repository guardrails and plugin versioning.

Provenance

Part of the operator-requested saga; independent-work scheduling explicitly included.
Extends the fleet bounds established by closed #44 without removing them.

Blocked by #344

Part of #343