Improve performance on big BN

[Pierre Denis]

Lea's exact algorithm does not scale well on some BN.

Here is an example using alarm.bif (37 nodes, 46 arcs, 509 parameters)

declare_namespace(globals())
alarm_bn = read_bif_file("alarm.bif", create_vars=True)

The previous statements run very fast (BN creation, without much calculations). Now, the following query:

HR.given(PAP=="LOW").calc()

takes 3 minutes and consumes more than 10Gb in memory!

This can be improved dramatically by avoiding to expand generator into tuple when building the final probability distribution.

[Pierre Denis]

Remove Alea._zip_vps method and avoid expanding generators into tuples (refs #90)

→ <<cset cbc8a3dee089>>

[Pierre Denis]

After the fix, the processing time is reduced to about 30 seconds, and the memory consumption is unnoticeable.