If a string's contents are fully contained in the other string, only
treat the second as redundant if the first string has at least four
words.

If we set the window/dialog as the locus of focus and then immediately
get a focus-change event, the presentation of the window/dialog gets
cut off.

Chromium-based browsers appear to be applying name-calculation to certain
divs which are not focusable. This causes us to double-present information,
namely speaking the div name (which came from its contents) followed by the
current line (i.e. the contents). Therefore if the section is not focusable,
don't include its name.

In web content a word can be made up of several accessible objects.
This is not the case in text inputs, so don't include any previous or
next objects when getting the word contents at offset.

When `priorObj` is not explicitly set, the current focus is used. This makes
sense when we're about to present something that will become the focused
object. However, if we had to update focus prior to presentation, this can
cause us to conclude incorrectly that focus didn't actually change. And when
we think focus didn't change (e.g. arrowing from one line of a heading to
another line in that same heading), we try to be less chatty.

locus_of_focus_changed() was sometimes setting `priorObj` to `old_focus`.
That should be the default, and we should special-case (and document) any
instances where doing so is not desired.

See issue #541.

Most sections are excluded from ancestor generation because they are
deemed layout-only. But if an ancestor section fails the is-layout-only
test (e.g. due to having a name or being focusable), we will attempt to
generate speech for it. This can result in some content being presented
twice. Therefore, when generating speech for an ancestor section, limit
the generation to the bare minimum and return before we do any text-related
generation.

It's printMessage() in v47; print_message in v48.