check_restart with a ignore fails pahse

### Proposal
Actutally if we define a check_restart in a service or check definition, the only way to overlap the start or restart phase of an allocation is to set a grace time.

This works so far so good but makes the allocation unavailiable during this grace time. Escpacially this increase the boot time of the cluster.

Is it possible to make checks during the grace time as defined in checks' interval defintion and ignore failures but mark the service as available if the check is positive during the grace time ?

Start by tracing how check_restart, service or check definitions, grace time, and check intervals interact during an allocation's start or restart phase. Clarify the desired state transitions and failure-handling rules, then verify that checks during grace time can mark a service available without making the allocation unavailable when a check fails.

Written by the indexing model from the issue text.