Verifying a time-sliced waveform loader: hash exact tiled windows against a 1-worker whole load

When a signal history is loaded as N time slices and concatenated (with normalization: last value per time, repeated values dropped), verify by (1) cutting slices exactly AT a change time, at t-1 and t+1, as a 1-unit slice, before the first change and after the last, then comparing the raw record buffer to one whole load; (2) through the protocol, hash exact windows tiled so each has <= 2*buckets changes (split recursively when not exact) and compare with a reference build that never splits. Also check: Promise.all over per-signal jobs makes the first rejection trigger fallback whole loads for the others while their jobs still run (double load and storage); use allSettled. Cancel-to-background designs need a way (free/close) to drop jobs nobody waits for, or the pool keeps working on removed rows.