Bug fix in parallel_scan

[Sudha247]

Fixes bugs reported in #38 and #59. The fixes are:

1. Wrong entry in result: The final entry in the array was wrong when ~num_additional_domains:1. Caused by pool size 2 being omitted while accounting for offsets, fixed in e03ed86

2. Rejecting array size less than pool size: add a fast path in parallel scan #39 also tries to fix this by calling a sequential scan when pool size is greater than length of the input array. As @nilsbecker pointed out, this may not be the most efficient way for long running operations. What's done here is instead assuming the scan is going to be run on n domains, where n is the size of the array. So, the pool size is minimum of number of domains available and length of array. So each operation runs on a domain of its own when number of domains is greater than array length. (fixed in c8bdc24).

Also adds @jmid's test in #59.

[kayceesrk]

@jmid If this looks good to you, let me know (or mark as approved) and I can merge this one..

[jmid]

LGTM! 👍‍
I've also run a few local tests on 4.12.0+domains and a recent 4.14.0+domains with these on top of #50 and it works nicely.
Thanks @Sudha247!