Add sparse sum helper to util

[stress-tess]

Add sparse sum helper to util. This strips out the extra work from the sum aggregation, but still uses radix sort to merge the 2 sorted indices list, which is definitely overkill. The next step is to optimize the merging

[stress-tess]

Here is the code I used to test that this matches what we get when doing a groupby over the indices and sum aggregation

>>> def tester(n=5):
   ...:     import pandas as pd
   ...:     select_from = ak.arange(10**n)
   ...:     inds1 = select_from[ak.randint(0, 10, 10**n) % 3 == 0]
   ...:     inds2 = select_from[ak.randint(0, 10, 10**n) % 3 == 0]
   ...:     vals2 = ak.arange(10**n, 2*10**n)[inds2]
   ...:     print("started sumHelp way")
   ...:     timer1 = pd.Timestamp.now()
   ...:     helper_idx, helper_vals = ak.util.sparse_sum_help(inds1, inds2, inds1, vals2)
   ...:     timer2 = pd.Timestamp.now()
   ...:     print("finshed sumHelp way, took:")
   ...:     print(timer2 - timer1)
   ...:     vals = ak.concatenate((inds1, vals2))
   ...:     print("started gb way")
   ...:     timer1 = pd.Timestamp.now()
   ...:     gb_idx, gb_vals = ak.GroupBy(ak.concatenate([inds1, inds2])).sum(vals)
   ...:     timer2 = pd.Timestamp.now()
   ...:     print("finshed gb way, took:")
   ...:     print(timer2 - timer1)
   ...:     print()
   ...:     print((gb_idx == helper_idx).all())
   ...:     print((gb_vals == helper_vals).all())

[jaketrookman]

Looks good to me