Pass heap tuple counts to parallel vacuum as double

Hi,

I noticed that parallel_vacuum_bulkdel_all_indexes() and
parallel_vacuum_cleanup_all_indexes() take the heap tuple count as long,
while lazy vacuum passes a double and vacuumparallel.c stores it as a
double in shared memory for the index AMs. Where long is 32 bits, as on
Windows, a table with more than about 2.1 billion tuples gets a wrong
num_heap_tuples only when its indexes are vacuumed in parallel. It's
rare in practice, but the serial path has no such limit.

The attached patch changes the parameter to double.

Thoughts?

Thanks,
Baji Shaik