Double to int casting can cause precision issues

[GiacomoBoldrini]

When computing bin sums following Barlow-Beeston approach, a double -> int -> double cast is made to obtain the number of effective events here and here. In presence of large number of events in a bin with small associated error the values can easily exceed int 32 bits giving errors as reported

------------------------------------------------------------
210        5058.074707     0.084911        total sum                     
210        5058.074707     0.084911        excluding marked processes    
210        -2147483648.000000 -nan            Unweighted events, alpha=-0.000002
  => Number of weighted events is below poisson threshold
    sm_lin_quad_cHj1     5137.480469     0.081002                                      
                         -2147483648.000000 -nan            Unweighted events, alpha=-0.000002
Error in <ROOT::Math::Cephes::igami>: Wrong domain for parameter a (must be > 0)
Error in <ROOT::Math::Cephes::igami>: Wrong domain for parameter a (must be > 0)
      => Product of prop_bininclusive_all_bin210_sm_lin_quad_cHj1[-2147483648.00,0.00,0.00] and const [-0.0000] to be poisson constrained

Simply changing int to int64_t in both lines seems to solve the problem

210        5058.074707     0.084911        total sum                     
210        5058.074707     0.084911        excluding marked processes    
210        3548512580.000000 59569.392980    Unweighted events, alpha=0.000001
  => Total parameter prop_bininclusive_all_bin210[0.00,-7.00,7.00] to be gaussian constrained