improve(dashboard): describe both drain-soonest modes honestly in routing settings

The routing settings help text still describes the sticky mode with its
original promise — at every fresh selection it prefers the account whose
weekly window resets soonest — which production tracing disproved: with
several concurrently active sessions the most recently started session
holds all traffic and fresh selections effectively never happen.

Anyone choosing a strategy on the /settings page reads that text and
picks the sticky mode expecting drain behaviour it does not deliver, and
finds no explanation of how the strict variant differs.

Now the text states the sticky mode's actual behaviour (never preempted;
ranking only orders re-selection and failover), describes the strict
mode as ranking every selection by earliest weekly reset with the model
family capacity skip, and the description test pins both statements
instead of the retired tiebreaker phrasing.