perf(client): make getTrades pagination O(N) instead of O(N^2)

[KENILSHAHH]

Summary

• ClobClient.getTrades accumulated paginated results with results = [...results, ...response.data] inside the while loop, re-allocating and copying the entire accumulator on
  every page — O(N²) total work in trade count, plus heavy GC churn.
• Switched to in-place results.push(trade) appends. Same semantics, now O(N).
• Added a guard for empty/undefined response.data pages (defensive; matches original behavior).

Context

For accounts with large trade histories (e.g. 50 pages × 500 trades = 25k), the old code
performed ~1275× more element copies than necessary and allocated a fresh array per page. This
was the only hot paginated loop with the issue — getTradesPaginated and getBuilderTrades
already use single-page returns.

---

Note

Low Risk
Low risk: change is limited to how getTrades accumulates paginated results and adds defensive handling for missing page data, with new tests covering behavior and performance.

Overview
ClobClient.getTrades now accumulates paginated trades via in-place appends (push) instead of rebuilding arrays each page, eliminating quadratic copying/GC overhead on large histories.

Adds a guard to skip empty/undefined response.data pages, and introduces a vitest suite that mocks paging to verify ordering, cursor/only_first_page behavior, and includes a loose performance regression test to catch a return to O(N^2).

^{Reviewed by Cursor Bugbot for commit c17cc40. Bugbot is set up for automated code reviews on this repo. Configure here.}

[KENILSHAHH]

@poly-rodr , @arun-poly any chance to get this reviewed ?

[arun-poly]

@poly-rodr , @arun-poly any chance to get this reviewed ?

@Pradeep-selva can review this pls