dbconsole: create a top level (dissmissable) warning banner around cluster misconfigurations

Create an generic framework that db engineers can use to create dismissible warnings that will then surface in the dbconsole or some other downstream alerting api.

Some example use cases for this:
- the customer exceeds some documented recommendation, but that we do not feel comfortable implementing a hard guardrail for-- e.g. the 10k descriptor limit.
- the customer tunes a cluster setting to some strange value, but that we are not comfortable putting a hard guardrail on. This happened last night on the 300 node test last night, accidentally.
- as another example: in a support issue, we ask the customer to temporarily tune some cluster setting, but the customer then forgets to reset the cluster setting after incident is over.
- we detect the customer using a crdb_internal query or querying the system database directly (i realize we are in the process of putting explicit guardrails here)

Jira issue: CRDB-51580

Start by mapping the dbconsole and downstream alerting API mentioned in the issue, then review how cluster configuration and customer usage are currently detected. Clarify the generic warning framework's scope, dismissal behavior, and consumers; done means documented misconfigurations can produce dismissible warnings in dbconsole or the downstream API.

Written by the indexing model from the issue text.