Levenberg-Marquardt fitting for sensor calibration

[NicolasM0]

This PR aims to add a new Levenberg-Marquardt fitting for sensor calibration. It optimizes scales and offsets (without off-diagonals). It could be used for accel calibration as discussed here #18166.

As a first step, the existing code is factorized.
and then a new fitting ("ortogonal_ellipsoid") is added.

This is a draft and it can certainly be improved. I am not sure how to make it cleaner. I could define an "algorithm" class from which each specific algorithm can inherit.

Not tested much for now but I have done 3 mag calibrations: one with the actual code, one with my new implementation (should have the same results but as input data are not the same...), and one with the new orthogonal_ellipsoid method (not aimed to be used for mag calibration but just to see if it works).

	actual	new	orthogonal_ellipsoid
CAL_MAG0_XODIAG	0.001	-0.003	0.000
CAL_MAG0_XOFF	0.017	0.012	0.014
CAL_MAG0_XSCALE	1.011	0.999	1.003
CAL_MAG0_YODIAG	-0.014	0.001	0.000
CAL_MAG0_YOFF	-0.042	-0.043	-0.051
CAL_MAG0_YSCALE	0.981	0.983	0.974
CAL_MAG0_ZODIAG	0.088	0.078	0.000
CAL_MAG0_ZOFF	-0.179	-0.187	-0.178
CAL_MAG0_ZSCALE	0.984	1.042	1.016

[dagar]

Once this settles I'll go back and update/finish #18166.