Additional inference rules for nonvisual reading

[HadrienGardeur]

Once support for parsing HTML/CSS (#180) has been added, we should be in a position to:

• check for decorative images (with an ARIA role set to presentation or none, which should translate into the presentation role in Guided Navigation)
• identify images where the height/width ratio makes them decorative
• and/or images with decent enough alternate descriptions (rules still TBD, but for example we'll check for filenames using a regexp)

If all images are:

• decorative
• with a decent enough alternate description
• ignored based on the work done in #179
• or identified as a cover

… then we should be able to infer "displayTransformability": "textual".

Given the need to parse HTML to support this inference, we'll put this feature behind an optional flag for both rwp manifest and rwp serve.