[1.0] Add run cancellation and global pause of new starts

Operators need a bounded control surface for one run or new process-wide starts, without claiming complete cancellation or durable resumption of a correlated workflow.

Scope

• Expose the same authorized operations through Python, HTTP and dashboard: cancel a selected run, pause new starts globally, and release that pause.
• While paused, already-running work may finish and all new/held input remains subject to admission limits.
• Show cancellation requested until termination is confirmed. Require permission for the entire selected run, including mixed-origin inputs.

Acceptance criteria

• Cancelling one of two concurrent runs does not cancel the other or its unrelated successors.
• Global pause prevents new starts; release starts held eligible work once without exceeding admission or concurrency limits.
• Existing outputs and external side effects are not portrayed as undone; process-local pause is clearly labeled and no cross-restart resume guarantee is made.

Boundaries

Full correlation-scoped pause/resume/cancel, durable control decisions and shared-cascade ownership are deferred under #294.

References

• Parent topic: #294.
• Implementation dependencies: #430, #433, #440, #445.
• src/flock/components/server/control/control_routes_component.py
• src/flock/orchestrator/scheduler.py
• src/flock/core/orchestrator.py

Start by reading src/flock/components/server/control/control_routes_component.py, src/flock/orchestrator/scheduler.py, and src/flock/core/orchestrator.py, then trace the implementation dependencies #430, #433, #440, and #445. Verify the authorized operations across Python, HTTP, and the dashboard, with cancellation and global pause behavior matching the listed acceptance criteria and boundaries.