smoothMin and smoothMax documentation addition (see also buildings-issue #3262)

[RouvenZ]

Hello all, hello @EttoreZ,

as requested in the buildings-library issue #3262, I opened the issue here as well (a bit belatedly, please excuse!).

As mentioned, for some implementations it is necessary to cap values (say a variable x) smoothly (over an interval dx) and additionally to not have them move out of bounds, especially not above (for smoothMin) or below the cap s (for smoothMax).

The smoothness requirement for regStep introduces an overshooting (for smoothMin) or undershooting behaviour (for smoothMax), making it necessary for the cap and the higher/lower bound for a value to differ. While there is a remark within the docu for Examples.SmoothMin, that this can happen, it is not quantified there.

However, it is possible to quantify exactly at what value x the over-/undershooting happens and to what extent.

For smoothMin, the maximum overshoot happens at $x = s - \frac{1}{2} dx \cdot (1 - \sqrt{3})$ at a function value $f(x) = s + \frac{3}{16} (2 \sqrt{3}-3) \cdot dx \approx s + 0.09 \cdot dx$.
For smoothMax, the maximum overshoot happens at $x = s + \frac{1}{2} dx \cdot (1 - \sqrt{3})$ at a function value $f(x) = s - \frac{3}{16} (2 \sqrt{3}-3) \cdot dx \approx s - 0.09 \cdot dx$.

So I would suggest adding the following text snippets to the docu in order to not overcomplicate things and maintaining a safe distance to machine precision eps:
For smoothMin: If variable x1 is to be capped by its max-value s smoothly and without overshooting, write smoothMin(x1, s-0.1 * deltaX, deltaX).
For smoothMax: If variable x1 is to be capped by its min-value s smoothly and without undershooting, write smoothMin(x1, s+0.1 * deltaX, deltaX).

What do you think?

Thank you all for the amazing work on this library and have a nice day!

Cheers
Rouven