Change storage type for SF histograms used by the efficiency helpers

[cippy]

The helpers are fed with boost histograms created with storage = hist.storage.Weight().
It can be changed to storage = hist.storage.Double() because we don't use the bin errors explicitly anywhere for the smoothed efficiencies, the uncertainties are stored as independent layers of the histogram using the variation axis (the binned case is different, but that one is handle by a different set of python/C++ scripts).

This would reduce the size of these histograms by 2, potentially improving running performance.
It also requires changing the C++ implementation of the helpers, since when reading with h->at(cell) the returned object will be a double and no longer a duplet with values and variances.