Some cases come up when we do pageserver restarts/migrations while LSN leases are in play, and the pageserver's gc_cutoff has advanced past the lease.

1. Validation of lease LSN vs. gc_cutoff: this makes sense for first request but not renewals. We cannot know the difference from the server side, so: page_service lease requests should not do the validation (these are used for renewals), but http API requests should do the validation (only used for initial grant of lease).
2. getpage requests are validated against latest_gc_cutoff, with a special case for requests at leases. However, after pageserver restart/migration, there is a period where the compute may send getpage requests before it has sent its lease requests. To narrow this window, we should make compute_ctl send a lease renewal as soon as it sees a /configure API calls that updates pageserver_connstr.

On point 2, the suggested change doesn't eliminate the case of bad requests, but it limits to:

• A very short period after migration/restart
• Only happens when the GC cutoff has advanced past the lease location: most uses of static endpoints on tenants with ~24hr PITR will never hit this case.

We must somehow document that this is a legitimate case where we might see "client requested an LSN that has been GC'd", so that we don't get too worried if we ever see this