perform scaling of the object step length for LSTSQ multislice in a similar way as FoldSlice

[a4894z]

In FoldSlice, when computing LSTSQ multislice steplengths (using either the joint probe/object steplength computation or the non-joint), they divide the computed object step length by the number of slices and the number of probe modes.

I include this scaling in LSTSQ multislice for this pull request.

[mdw771]

step_size already does this so we don't need a separate parameter. I will revert this.

[mdw771]

Changes

Changes are made to match PtychoShelves in the following discrepancies found between the two packages:

Item	Matched/fixed?
Initialize probe modes using Hermite polynomials.	Y
Normalize probe magnitudes using max power of DPs.	Y
Always regenerate secondary incoherent modes and OPR eigenmodes even when initial probe is given	N
Auto center probe	Y
Far-field propagator FFT uses "forward" instead of "ortho"	Y
Scale probe magnitude by 1/sqrt(N) to compensate for FFT factor	N
Orthogonalize incoherent modes at the beginning	Y
Only correct probe intensity scaling during the first epoch	Y
Build object preconditioner with only the principal mode and divide preconditioner by 2	Y
Calculate object step size with only the principal probe mode; use psi_{i-1} for intermediate slices (Pty-Chi had a mistake)	Y
Calculate probe step size using preconditioned probe update	Y
Pty-Chi has a mistake in mode slicing in variable probe update	Y
Pty-Chi has a mistake in calculating the projection of the probe update residue to the current eigenmode	Y
Pty-Chi has a mistake where the chi used for OPR mode/weight update is modulated by the object	Y
Optional multimodal update; if false, calculate object update with only probe mode 1; if true, use all the modes, and scale object step size by the number of probe modes	Y
Momentum acceleration uses Fourier error to determine whether to apply momentum or not	Y
Pty-Chi has a mistake in correlation calculation for momentum acceleration where the norm wasn't calculated correctly	Y

For the most updated list of discrepancies, see https://git.aps.anl.gov/ptycho_software/pty-chi/-/issues/32.

API changes

• LSQMLReconstructorOptions.solve_step_sizes_only_using_first_probe_mode default changed to True
• LSQMLReconstructorOptions.momentum_acceleration_gain added
• LSQMLObjectOptions.solved_step_size_upper_bound removed
• LSQMLObjectOptions.optimal_step_size_scaler added
• LSQMLObjectOptions.multimodal_update added
• LSQMLProbeOptions.solved_step_size_upper_bound removed
• LSQMLProbeOptions.optimal_step_size_scaler added
• LSQMLProbeOptions.eigenmode_update_relaxation default changed to 1.0 (was 0.1)
• LSQMLOPRModeWeightsOptions.update_relaxation default changed to 1.0 (was 0.1)