Fix usage pagination and billing for failed artist scrape starts
Equal-timestamp usage charges can disappear between pages because the old time cursor omitted the row ID, and artist-wide scraping charged for failed starts. Usage now pages by timestamp plus ID without losing database microseconds, and scrape billing counts only returned run IDs.
Legacy ISO time cursors remain accepted as strict time boundaries; callers should start a new traversal to cover timestamp ties. Cost/ID ordering and the independent period totals remain unchanged. Coordinated Marketing/Docs changes publish the opaque cursor example, nullable response fields, and correct micro-dollar scrape prices.
Validation: Final-head CI passed 6,407 tests (13 environment-dependent tests skipped); the local full suite passed 6,406 tests before the two added legacy/composite period-boundary cases, which also pass; lint and changed-file formatting pass. The production build passes with font-download network access. Local fixtures reproduce tied-row loss, microsecond truncation and charging failed starts before the fixes. No paid scrape, production ledger mutation or historical billing discrepancy is asserted.
Tracks recoupable/mono#345, recoupable/mono#346 and the provider-free checks for recoupable/mono#336. Publication and production readback follow coordinated approval; this PR does not close the epic's broader runtime gates.