storage controller: be more tolerant of a pageserver's startup time

[jcsp]

We mark a node offline after MAX_UNAVAILABLE_INTERVAL_DEFAULT (30 seconds) of failures to respond to heartbeats.

We should be more generous during startup: when a pageserver sends us a re-attach request, we should tip off the heartbeater to be more generous. Currently the pageserver's processing of the re-attach respond can be quite time consuming.

This is similar to the k8s distinction between a readiness check and a status check: we should be more tolerant when waiting for readiness during startup, than when checking for responsiveness during normal runtime.

(The actual init_tenant_mgr slowness is addressed in #7553, but this ticket still stands: we should be more tolerant during startup than we are during normal operation.)

[jcsp]

@VladLazar let's pick this up as part of the rolling restart work: we should flip the node into its more tolerant mode on:

• Notification that it is draining (or a separate hook for "I'm about to shut this down"?)
• Starting to handle re-attach