More number-crunching loops run several values at once, now on Windows and Intel Macs too, and faster than the reference browser

Earlier releases taught Floati to process several array elements at the same time on the chip's vector unit for loops that transform one array into another or fill an array from a formula. This release widens that to more of the loops real pages actually run: adding up a whole array into a running total, smoothing or blurring by blending each value with its neighbours, clamping values into a range with min and max, and taking square roots across an array. On this work Floati now finishes several times sooner, and faster than the reference browser, which still does it one value at a time. Just as important, these speedups now run on Windows PCs and Intel Macs, not only on Apple Silicon. On every machine Floati uses only the vector operations that processor is certain to support and quietly falls back to the ordinary one-at-a-time path for anything else, so the result is always correct and it never crashes on older hardware. It engages only when Floati can prove every read and write stays inside the array and that doing a batch at once gives exactly the same answer as one at a time, so the data is identical to the last bit. It all runs on your device.