Larger "small" heuristics for smaller max_width

[kaj]

I'm not sure where to put this. It is kind of related to #2299, #2447, and #2437. But I might be happier with a simpler solution than whats suggested there, so I open a specific issue.

I use max_width = 78 (since I want a git diff of my code to fit within an 80 chars wide terminal), and I want "short" struct literals and if-else statements to be single line, so I leave use_small_heuristics to its default true value.

Since most of the the default WidthHeuristics are defined in relation to max_width, which defaults to 100, I think many of the heuristics simply get too small when using a smaller max_width.

So my intent was to suggest non-linear values in WidthHeuristics::scale (maybe simply leaving it as is for max_width >= 90 and otherwise using e.g. values 85, 25, 50, and 75 instead of 60, 18, 35, and 50).

But maybe something like #2447 is a better alternative? Maybe it can be generalized to having different named set of heuristics? On the other hand, less configuration is better, so I rather like the concept that the heuristics should attempt to adapt as good as possible to the given max_width.

[topecongiro]

We will restore the per-item width configuration in 2.0.