Glossary: Improve performance of parsing translations for adding tooltips

[akirk]

Closes #1383.

This improves the implementation of #1359 for a better performance. The key improvement is shortening the regex for splitting the singular into chunks of text and glossary items.

Before, for example for Galician, it created a 20k regex of the format:

(\bauthentication\b)|(\bauthentications\b)|(\bauthenticationes\b)|(\bauthenticationed\b)|(\bauthenticationing\b)|(\bminification\b)|(\bminifications\b)|(\bminificationes\b)|(\bminificationed\b)|(\bminificationing\b)|(\bnotification\b)|(\bnotifications\b)|...

The new code improves it to a 6k regex like this:

\b(authentication(?:s|es|ed|ing)?|authenticate(?:s|es|ed|ing)?|...

Before, the regex performance was about 0.3s per preg_split() call, the new one is < 0.1s.

I also refactored the regex generation to happen inside the function cache it interally using the static keyword and also the generation of the reverse lookup to identify the assigned glossary term.

[pedro-mendonca]

Here are my results for a full WP 5.9 dev project with 5148 strings and a glossary of 271 entries.

500 strings per page

I had a 28 seconds page load
This PR reduced to 6 seconds

All the 5148 strings on a single page

I had 337 seconds page load
This PR reduced to 38 seconds

[ocean90]

Thanks for working on this! Definitely performs better than what we currently have.

I only noticed one minor difference when I have an entry for "comments" and "comment":

Before	After

The list is now inverted where the most relevant entry is now the last item.