Fix: speaker segmentation doesn't lose words

[jamesdools]

Describe what the PR does

This addresses the groupWordsInParagraphs function inside the Kaldi-to-draft editor.

Usually, we have speaker segmentation attached from Kaldi, so we haven't been calling this.

But it's throwing away any words that go beyond the last full stop (or exclamation etc.)

Eg. if you input: Live Kaldi Test. Here is some transcript data sorts to just `Live Kaldi Test.'

and if you input `Here is some transcript data, it results in an empty list of paragraphs (which bricks the whole app 🙄😄)

Super quick PR. Added some tests too - let me know if you need more context.