As mentioned in the code review on #94998, perform the following optimizations:

• When encountering that the cache is stale in GetInstance(), refresh the cache.
• In multtenant-upgrade, disable distSQL instead of removing the check to validate that tenant 11 is still able to query the database.
• Build a version of GetAllInstancesNoCache() which accepts a sessionID and only looks for that specific instance.
• Disable the RPC circuit breaker throughout TestTenantUpgradeInterlock

Jira issue: CRDB-25560

Epic CRDB-31933