refactor: drop dead model/limit plumbing and unused render helper

resolveSearchDefaults no longer accepts a per-call model override and
SearchParams no longer carries limit; neither was reachable from the
tool schema (limit is applied client-side in formatForLLM). Also fold
asPositiveNumber into asPositiveInteger and simplify recency matching.

test/e2e-models.test.ts

@@ -51,7 +51,6 @@ describe("Perplexity model selection e2e", () => {
             query: "Say exactly OK",
             model,
             incognito: true,
-            limit: 1,
           },
           token,
         );