* stream_utils: Before coordinate conversions to observational longitude
assumed two (mass) categories. Now allow arbitrary number of categories.

* stream_utils: Before computing and loading data was carried out by the same
procedure. Now split `save_obs_Qrho` to compute coordinate conversion and save
data, and load_data_Qrho to load data.

Before shot-noise of Qrho was computed as sigma = err * rho. Now it is
sigma = err.

Instead of interpolating PDF(PDF(Qrho)) with a gamma, multiply
directly the histograms.

The exponential distribution for the bins on the PDF tail is due to
poor binning resolution. Changing linear binning to log show lognormal
distribution instead of exponential.

* Integrate over log range.

* Add Heaviside cutoff to priors.