New inference rule for "accessMode": "visual" #89

With our current inference rule, we infer "accessMode": "visual" when:

• a publication does not contain any metadata for accessMode
• and we detect at least one image in readingOrder/resources that's not a cover

As a follow-up to our discussion on #87, we want to replace this with a new one:

• this new inference would exclusively be triggered when inspecting HTML resources as well or when the publication has images in the readingOrder
• it would be triggered if the publication has at least one image that's neither decorative nor a cover

We could handle this task after #88 since we'll have all of the necessary info for that.