Clipping of step-size changes of CSA

[nikohansen]

Currently, the step-size change in CSA, $\exp\left(\frac{c}{d}(||p_\sigma|| / \chi_n- 1)\right)$, is a factor clipped into $1/e$ and $e$ via the hard coded attribute .max_delta_log_sigma = 1. This is however undocumented!

The option

{'CSA_clip_length_value': 'None  #v poorly tested, [0, 0] means const length N**0.5, [-1, 1] allows a variation of +- N/(N+2), etc.'}

is related and restricting the size of a single summand in $p_\sigma$.

Arguably, step-size clipping is universal and could or should be handled independently of the control method?