OFFSET_T,P fit

[julienguy]

OFFSET_T,P are calibration parameters defined by the relation EXT_T,P = INT_T,P + OFFSET_T,P , where INT_T,P are the "intrinsic" angles in the frame of the positioners ; and EXT_T,P represent angles in the frame of the petal. The issue is that they serve 2 different purposes in the current calibration code:

1. OFFSET_T,P define the angles of the hard stops in the petal frame because the rehoming procedure attributes fixed INT_T,P angles for the hard stops of the positioners.
2. OFFSET_T,P are used in a calibration fit as zero points, or constant offsets, between the "measured" angles and the "expected" angles INT_T,P. As such, they compensate for any drift between the true angles of the positioners and the parameters INT_T,P. Indeed INT_T,P are not measurements, they are values initialized after rehoming the positioners, and then incremented in the code for each move, with their values saved in the DB. The problem is that the INT_T,P values do not perfectly track the true angles, errors get accumulated, and the values of INT_T,P have to be recalibrated from time to time.

So the calibration fit should always fit for offsets, but distinguish whether to attribute this measured offset to a measurement of the actual angle of the hard stops (i.e. OFFSET_T,P), or a correction to INT_T,P.
The former case should be considered only when the calibration includes a rehoming of the positioners, AND if it is expected that the drifts of INT_T,P would be negligible (in particular for a non-linear response, the situation is more complicated).

Applying corrections to OFFSET_T,P in a calibration sequence that does not include rehoming, or when many moves have been performed since rehoming so would lead to incorrect angles for the hard stops.