test(reminders): stabilize Azure Table reminder tests

[ReubenBond]

Problem

Azure Table reminder tests still had three independent race conditions after the broader reminder-clock work: the restart test waited for one post-reset tick while asserting two, the cross-grain test let independent waits advance one shared fake clock, and failover tests resumed after best-effort liveness stabilization without proving that the shared client could route grain calls through the replacement gateways.

Existing repairs reviewed

Recent merged work addressed adjacent reminder races: #9989, #9997, #10033, #10159, #10296, #10312, #10317, #10400, #10421, and #10430. Current main still skips both #9557 methods and retains the mismatched restart wait and independent cross-grain clock drivers. The latest stale-gateway failure occurred on a head containing #10400; #10430 merged afterward but only changes exact-due reminder recovery and does not affect the failing client GetCounter route.

Solution

Use the single reminder-clock orchestrator and exact counter assertions for the restart and cross-grain scenarios, then re-enable both quarantined tests. Add an explicit grain-reachability phase before and after failover: only expected lifecycle transport failures are retried, each retry re-runs liveness stabilization, and the existing test cancellation deadline remains authoritative.

This preserves the failover coverage without sleeps, test skips, or increased timeouts, and makes failures report the affected grain set and last transport error.

Fixes #9557