Fix some import module completions being dropped (and fix flaky test too)

[eddiemundo]

Fixes #2577 and maybe other issues involving import module completions.

The superficial reason why the module completions test was failing is that HashMap's toList ordering is not deterministic between runs. This meant that the list ordering of the target files/modules would occur in different orders.

Of course why should the order of module names affect anything with regards to completions?

The test generates 3 targets, 2 TargetModules and 1 TargetFile. Since we seem to only care about actual modules we convert TargetFiles to empty strings. This list of module names (including the empty string) eventually get passed to the fuzzy matcher which attempts to run the matcher on the names in parallel.

Interestingly different orders can cause the parallel strategy to give different results which should be impossible unless the Strategy changes its input. Therefore the strategy does change its input.

Turns out the function used to chunk vectors had an off-by-one error when doing the Vector chunking, and if the empty string wasn't in the last position some module completions would be dropped.
e.g.

> chunk 3 [1,2,3]
[[1,2], []]

Also interesting is the comment on the chunkVector function seems to have come from eval plugin usage and has the correct output underneath it, yet when I eventually ran the eval plugin for kicks it produced the wrong output 😅

The fix is just adding +1 to the slice length.

[jneira]

Wow, good catch, many thanks for the fix

[pepeiborra]

Thanks for the fix @eddiemundo awesome stuff. The lack of unit tests in this code (I authored it) is disturbing...