Improve performance between lambda changes

[jxjj]

Random thoughts for later:

• Once calculated, save the singular value decomposition (svd) of the basis matrix, U, sigmaVector,V and maybe U'*y. Then we should be able to recalc betas more quickly.
• Provide an function for updating lambda.

Sticking with a functional approach, maybe the spline object returned from a smoothing spline, also includes _meta information for recalculating and an update or with function (or some other word that might be more suggestive that a new spline is returned rather than the original spline is mutated?)

const spline = await simpleSmoothingSpline(data, { lambda: 1 });
// first run so a bit slower, but work is saved to _meta

const newSpline = await spline.update({ lambda: 2 })
// faster since we can reuse previous work

Where spline might look like:

{
	points, // spline points
	fn,     // function to generate spline y for given x
	_meta: {
		type: 'smoothing'
		work: {
			svd,
			Utranspose,
			y,
		}
	update, // can use _meta.work to update and return a new spline object
}

Or, rather than moving state around in objects, move (back) to a OO implementation?

Possible OO API:

const spline = new SimpleSmoothingSpline(data, { lambda: 1 });
const points = await spline.getPoints();
// first run so a bit slower, but work is saved

const newPoints = await spline.update({ lambda: 1000 }).getPoints();
// faster since we can reuse previous work