Further computation steps work

[MelSumner]

From: https://github.com/w3c/accname/pull/122#pullrequestreview-646636771

• [ ] Consider adding examples (obscure/unrealistic as they may be) for how the actual behavior would change
• [ ] We explicitly say the embedded control stuff should take precedence over aria-label. That's now even clearer with the reordering. Should aria-labelledby (2B) be any different? Figure out if we're saying that aria-labelledby takes precedence over the value even for a label-embedded control, which seems wrong and neither Firefox nor Chrome do it:

data:text/html,<label><input type="checkbox">Flash the screen <input type="text" value="5" aria-labelledby="blah"> times</label><div id="blah">Blah</div>

• [ ] Consider if it makes sense to move the old 2E (now 2C) to 2B... So we'd have:

New 2B) If from descendant recursion and labelling another widget, use value if available New 2C) Otherwise, traverse aria-labelledby if available New 2D) Otherwise, use non-empty aria-label if available New 2E) Otherwise, use native markup if available