Review keywords

Find the words customers mention most often in a batch of product reviews.

Each review is turned into keywords like this, in order:

1. Lowercase it and cut it into words: maximal runs of the letters a-z. Everything else (spaces, digits, punctuation, apostrophes, hyphens) only separates words, so "don't" gives don and t.
2. Drop every word shorter than 3 letters.
3. Drop every word that is in stop_words (a list of lowercase words). This check uses the word as written, before step 4.
4. Reduce what is left to its stem using the first rule that applies, and no other:

• it ends in ies and has at least 5 letters: replace ies with y (cries becomes cry);
• it ends in ing and has at least 6 letters: remove ing (charging becomes charg);
• it ends in s, does not end in ss, and has at least 4 letters: remove the s (lasts becomes last, glass and bus stay).

A stem counts once per review, however many words in that review reduce to it. A stem's count is the number of reviews it appears in.

Input: reviews, a list of strings; stop_words, a list of lowercase words; and k, an integer.

Output: a list of at most k strings "stem count", sorted by count descending and, for equal counts, by stem in ascending alphabetical order. If there are fewer than k distinct stems, return all of them; if there are none, return [].