Expose coordinator startup as staged boot outputs

Coordinator.Start remains a large straight-line startup subsystem hidden behind one outer boot component. The server graph cannot see when useful coordinator capabilities become available, so unrelated work still waits behind maintenance phases that may each run up to their two-minute bound.

Break coordinator startup into a small number of cohesive stages whose typed outputs can participate in the existing outer boot graph. Model meaningful capability boundaries, not one node per controller, and avoid introducing a second nested graph.

Done when:

• The major coordinator startup phases and their consumers are explicit.
• The outer graph can start consumers as soon as their required coordinator capability is ready.
• Maintenance work gates only the consumers that actually need its result.
• Startup and shutdown ownership remain clear.
• Targeted tests prove that delaying one coordinator stage does not block an unrelated branch.