fix(discovery-core): bound /v1/history pages by wall-clock time

[Yoni-Starkware]

On mainnet, /v1/history pages for accounts with real notes ran past the 30s GCLB timeout (backend_timeout → 502). The I/O budget limits how many RPC calls a page makes, not how long they take. A key-filtered getEvents covers 81,920 blocks per pathfinder page at 2–3s, so the default budget of 10,000 (about 1M blocks) already takes 25–40s.

• IoBudget can carry an optional deadline.
• The gap scan fetches in 65,536-block sub-windows and stops at the first sub-window boundary after the deadline. The page also stops after a note block once the deadline has passed. Both checks run only after the cursor has moved, so the no-progress error path is unaffected.
• New limits.history_time_limit setting (HISTORY_TIME_LIMIT_SECS), default 10s. Documented in the specs and the deploy docs.

Tests: 3 new ones (budget deadline, gap scan, note step). Each fails when its check is removed. cargo fmt --check, clippy --all-targets, and cargo test pass for discovery-core and discovery-service. E2E not run.

🤖 Generated with Claude Code

---

This change is