Python seems to support 64 bits but vex will automaticaly discard extra bits and convert your attribute to 32 bits.
If you have a small point cloud it can be helpfull to do it with python, but if you need to manipulate a big number of points you will be in trouble (performances).
One way to reduce 64 bits manipulation can be to try to define an array of ids in 64bits at the detail level and use a basic index at point level referencing the detail array attribute.
At the end of your process and depending of your number of points you may want to create the final 64 bit attribute using python with the correct id using the index and the array of ids.
If you are using houdini engine, you may be able to do that operation on your engine instead of houdini.
Or you can pray that there is another solution or sidefx will one day fix that
Hope it helps...
From the documentation :
VEX uses 32 bit integers. If you use the AttribCast SOP to cast a geometry attribute to 64 bits, VEX will silently discard the extra bits if you manipulate the attribute in VEX code.
You can use underscores to break up long numbers.