new option per_r_par for covariance smoothing

[julienguy]

• add option per_r_par to picca.utils.smooth_cov
  if per_r_par=True (False by default), the correlation coefficients are averaged per (rpar,delta_rpar,delta_rperp) and not per (delta_rpar,delta_rperp).
• add option --smooth-per-r-par to picca_export.py

This is useful for some correlation functions where the default scheme is not accurate and can lead to non positive matrices.
For instance SIV and CIV side bands auto-correlations, or the lyb x lya cross-correlation.

Example:

# default (non positive output matrix)
picca_export.py --data /global/cfs/cdirs/desicollab/science/lya/y1-kp6/iron-tests/correlations/correlation-lyalyb-1-0-0/cf_lya_x_lyb.fits --out cf_lya_x_lyb_exp.fits --blind-corr-type lyaxlyb

# with new option (positive output matrix)
picca_export.py --data /global/cfs/cdirs/desicollab/science/lya/y1-kp6/iron-tests/correlations/correlation-lyalyb-1-0-0/cf_lya_x_lyb.fits --out cf_lya_x_lyb_exp_bis.fits --blind-corr-type lyaxlyb --smooth-per-r-par

[iprafols]

I have a couple of minor suggestions

[julienguy]

done