Describe the problem

The code still currently contains "imperfect" code paths that ought to be either generalized (to become independent of whether they run in system tenant or secondary tenant); or be entirely specialized to the system tenant because they don't make sense in secondary tenants at all (for example if they pertain exclusively to operating the storage layer).

Eventually we want all the SQL layer to not use KV-level data structures directly. All should go through a KV API over RPC and checked via the capability subsystem.

This issue is a dependency of #54252.

To Reproduce

Search for the following:

• OptionalErr
• OptionalGossip
• OptionalNodeID
• OptionalNodesStatusServer
• TenantSQLDeprecatedWrapper
• UnsupportedWithMultiTenancy

All these need to be eliminated.

Jira issue: CRDB-26645

Epic CRDB-26091