Lack of real-time cost savings visualization for the routing feature in the chat UI.

### Problem

The OpenSquilla README leads with a benchmark table where the router cuts cost from $6.233 down to $0.688 over 25 PinchBench tasks, about a 9x improvement, which is a great
headline story. But a user actually using the chat REPL never sees that story, because the existing /cost command in opensquilla/cli/chat_cmd.py and the per-turn footer rendered
by UsageSummary.render in opensquilla/cli/repl/stream.py only print absolute numbers like "12,345 tok · $0.001234" with nothing to compare against. The project's headline value
therefore disappears at exactly the moment the user is most ready to see it: right after a turn completes, when they are looking at the cost line on their own screen.

### Proposed behavior

Since the router config already knows which model sits in the most expensive tier (T3) for the operator's current provider, the gateway has everything it needs to also report
what the same prompt would have cost if it had been sent straight to that top-tier model, and then surface the delta. Concretely, extend the done event with two new fields,
baseline_model and baseline_cost_estimate, computed by re-pricing the same input and output token counts at the T3 model's rate, and have UsageSummary.render append one extra
line such as "saved ~92% vs $0.0152 if routed straight to opus-4-7" whenever the baseline cost is meaningfully larger than the actual cost. The line is suppressed when the
operator runs router=disabled or already routes everythin...