Recalculation of var_lss in true continuum problem

[p-slash]

True continuum analysis recalculates var_lss from pixel statistics in the following lines. It is a necessary step in the current framework. However, it is also problematic when the statistics are low. Example below for negative var_lss in true continuum analysis.
Maybe we can introduce a fail-safe such as replace only if estimated var_lss is within 50% of given var_lss. Thoughts? @iprafols @Waelthus and people working on true continuum analysis.

picca/py/picca/delta_extraction/expected_fluxes/true_continuum.py

Lines 137 to 143 in 6f44a68

	# the might be some small changes in the var_lss compared to the read
	# values due to some smoothing of the forests
	# thus, we recompute it from the actual deltas
	self.compute_var_lss(forests)
	# note that this does not change the output deltas but might slightly
	# affect the mean continuum so we have to compute it after updating
	# var_lss

[iprafols]

seems reasonable to me