Fix dictionary errors in OpenSCAD 2021

The unit tests for libdict.scad were failing on OpenSCAD 2021, which was confusing and a bit sad. I think the solution is currently fixing the version of OpenSCAD to 2019.5 in the CI script, but this MR makes it work on 2021 (and 2019).

I have traced this down to the summation, in _check_errant_match. The problem is that the increment condition (i=i+1, count_nl=count_nl+non_list[i]) gets evaluated for i==len(non_list) which raises a warning. I guess this warning is new in the latest release.

I have implemented two solutions to this:

• The first is simple but hacky: I wrap the indexing in the increment of count_nl with a ternary operator, i.e. the increment becomes:

  i=i+1, count_nl=count_nl + (i<len(matches) ? non_list[i] : 0)

  This works, but I think it's confusing and likely to be broken by accident, not to mention making an already-long line even longer.
• The second fix is longer but nicer: I replace the list comprehension with a tail-recursive function that sums the list. That will be internally optimised by OpenSCAD to a for loop, so the speed difference should be slim to none. That's the final commit on this branch and is the solution I suggest we use.