fix(mcp): reword explore guidance as advisory, not a quota

The trailing explore note said 'Explore budget: N calls ... spend your remaining calls ... Synthesize once you've used N' — agents read it as a hard quota, stopped exploring early, and fell back to grep+Read (more tokens, less complete). Reword it as explicit advisory guidance: the server never rejects or rate-limits extra explore calls, and exploring further is encouraged when the response doesn't cover the flow.

The tool schema description carried the same quota language ('Budget: make at most N calls for this project') — the higher-salience text the agent reads before ever calling. Reword it identically: advisory only, NOT a quota.

Also sync the eval probe string (probe-suite-envelope.mjs) and pin the new wording in tests with anti-regression assertions (medium tier must contain 'NOT a quota' and never 'remaining calls'/'Synthesize once'/'make at most').

scripts/agent-eval/probe-suite-envelope.mjs

@@ -85,7 +85,7 @@ try {
       epilogueCut: text.includes('omitted for size'),
       sectionCut: text.includes('output truncated to budget'),
       notShown: text.includes('Not shown above'),
-      budgetNote: text.includes('**Explore budget:'),
+      budgetNote: text.includes('advisory only, NOT a quota'),
     });
   }
 } finally {