Consider changing algorithm of tag name locator strategy

[lutien]

At the moment, algorithm of tag name locator strategy is very strict with requiring to use getElementsByTagName method. That works fine with regular html elements, but doesn't work for shadow roots right now across the browsers (https://wpt.fyi/results/webdriver/tests/find_element_from_shadow_root/find.py?label=experimental&label=master&aligned). Implementation wise, we could use for shadow root (or for all kind of elements) querySelectorAll and filter out non tag name selectors.

[whimboo]

Actually a duplicated of https://github.com/w3c/webdriver/issues/1610.