Delete or prune StructTreeRoot for --force-ocr/--redo-ocr/--skip-text and post warnings

[jbarlow83]

If a document has StructTreeRoot it's even more likely we're dealing with a bona fide OCR + tagged document, but the user can still request --force-ocr. Although there may be tricky cases like a mixed content document of digital output + scanned. This has interaction with --pages as well.

First, we need to warn if this object is present (and covering all requested --pages?) since it's a stronger warning against OCR.

Currently using --force-ocr will leave an invalid StructTreeRoot full of pointers to deleted objects. If we're doing --force-ocr we should delete all objects in the tree on each processed page.

For --skip-text we ought to leave StructTreeRoot intact. Except in contrived cases, the StructTreeRoot will not reference any pages with known text.

For --redo-ocr we also need to discard all objects, because our new objects may not match the old ones.