Framework: Optimize decodeEntities to skip unnecessary decoding

[aduth]

This pull request seeks to replace the current decodeEntities implementation with a much simpler alternative, with the hopes of improving performance. decodeEntities is a top culprit in profiling page transitions in Calypso, often the top time-consuming function in Chrome Timeline reports:

Before	After

(See earlier implementation results)

Implementation notes:

With these changes, decodeEntities will no longer decode all entities, only those which we expect to have been encoded by the WordPress _wp_specialchars method. In other words, it should now mimic the behavior of wp_specialchars_decode.

There was previously a test that expected entities like Ñ (Ñ) to have been decoded for plugin authors, but in my real-world testing, I find that these entities are never encoded to begin with and continue to display correctly. You can verify this yourself with an example plugin which specifies an author name including both "&" and "Ñ" characters. On the plugin detail screen, the author name shows correctly as "Andrew Duthie Ñáð & Friends". cc @enejb re: 6123-gh-calypso-pre-oss

Another minor optimization inspired by the wp_specialchars_decode implementation is to skip replacement attempts if there are no entities in the text argument, detected by presence of a & character.

Edit: After further review, it was determined that we rely on full entity decoding in many places (Reader, post titles with user-entered entities such as •). The changes here are limited to the above-noted inspired & check optimization, and memoization of the title-generating selector which makes use of decodeEntities on every screen.

Testing instructions:

Verify that wherever content is rendered using REST API data where text would be encoded (e.g. post titles, assigned categories), that the text continues to display as decoded. This should include expected encoded values & < > ' " as well as other special characters that are never encoded Ñ á ð.

1. Navigate to the post editor
2. Select a site, if prompted
3. Enter title "Ñáð & Friends"
4. Save post
5. Return to posts listing
6. Note that draft title is rendered exactly as "Ñáð & Friends"

Open questions:

Is there anywhere in the codebase that we're expecting decodeEntities to decode more than just the entities that would have been encoded by WordPress core logic?

cc @blowery

[matticbot]

Test live: https://calypso.live/?branch=update/wp-entities-decode

[aduth]

Noting reference to original implementation, since I'll plan to rebase it to the void: f4161484955c363c1e1219320bb09727de265b6d

[blowery]

Beautiful. 🚢