blur2 with stride

[Fil]

demo https://observablehq.com/@d3/blur-stride-dev

The API will have to change: having the stride at the end after the optional ry is not the best solution.

There is no need to change blur1 to accept a stride parameter, since it is the same as using blur2 with width=stride, rx=0, ry=r. We can intercept this in the API if we want, for consistency.

[mbostock]

I took the liberty of rebasing it with the latest commit 65942a9. Strangely the Laplacian example is faster but the simpler blur is slower, so I wonder if further optimization is possible or if the circular buffer strategy is not a good idea? I wonder if you would be willing to implement a benchmark that runs in Node. That would help compare the possible approaches (even though Node/V8 is only one engine to consider).

[mbostock]

This doesn’t feel right to me. It lets you blur RGBA, yes, but you can’t e.g. just blur the R channel, and now blurh and asymmetric with blurv. We need to think through the interface here some more.

[mbostock]

Closing in favor of c889f65 in #252. Thanks for the help!