Returns calendar (run_id comes from GET /v1/backtests/{id})

Period over period returns as a grid: rows x columns plus a compounded row total, with best/worst cell and the green cell rate. Computed from the FULL equity curve, never paged, because a partial window would compute a different grid. The bucket ADAPTS: decades when the run is too long to show by month, months from half a year up, otherwise days, otherwise hours, and never finer than the bar interval. Read granularity before interpreting a cell. columns labels the cells left to right; totalLabel names the trailing total (DECADE/YTD/MTD/DAY). The run's first bucket measures from inception and is therefore partial. data is null when the run is too short to grid at any bucket.