Now TORQUE uses struct numa_node.cpu_offset and numa_node.num_cpus to describe
cpuIDs, which precondition the cpu list is continuous, the same as mems.
However, according to cpuset interface, cpu list like cpus=1,5-6,7-9,13 is
valid. At lease the layout topology of my server reads
cpus=0,4,8,12,16,20     mem=0
cpus=24,28,32,36,40,44  mem=1
cpus=3,7,11,15,19,23    mem=2
cpus=27,31,35,39,43,47  mem=3
cpus=2,6,10,14,18,22    mem=4
cpus=26,30,34,38,42,46  mem=5
cpus=1,5,9,13,17,21     mem=6
cpus=25,29,33,37,41,45  mem=7

So I make a patch which uses numa_node.cpu_list and numa_node.mem_list to
describe discontinuous cpulist and memlist. 

The best struct to describe cpulist is bitmap instead of int array, which can
avoid duplicate cpuIDs.