fix(Readability): treat S and BDI as phrasing content

[SAY-5]

PHRASING_ELEMS is missing the HTML5 phrasing elements S and BDI, so Readability treats them as block-level and splits paragraphs around them (e.g. <div>foo <s>bar</s> baz</div> becomes <p>foo </p><s>bar</s><p> baz</p>). This adds both to PHRASING_ELEMS with a regression test.

Towards #997. I left U out for now because adding it changes the extracted output of the hukumusume test page, so that one is better handled separately.

[gijsk]

Thanks for the patch.

The testcase is a bit strange. We would typically add a minimal testcase to the testcases folder instead, with source and expected output.

I also think that we should include u and update the testcases accordingly (generate-testcase.js can do this for you if you delete the expected.html file first and then run it with the testcase slug as an argument). Or at least - I would like to see what changes that causes, before deciding whether any of these changes make sense. In isolation it is hard to see what this is really accomplishing.

[SAY-5]

Good points, thanks. Dropped the ad-hoc unit test and added a proper testcase under test-pages/phrasing-elements-s-bdi-u (source + expected regenerated with generate-testcase.js). I also added U to PHRASING_ELEMS as you suggested.

The fixture makes the effect visible: without these tags in the list, the run gets shattered into separate

blocks with the // nodes stranded outside the paragraphs; with them included the whole run stays in one

. The second div in the source keeps a real block boundary so the grouping is actually exercised.