Add OCR option to skip pages with a usable text layer

### Requested feature

Some PDFs contain both image resources and a usable native text layer.

Today, with `do_ocr=True`, OCR region selection is driven by bitmap coverage. That is useful for scanned PDFs, but it can do unnecessary OCR on born-digital PDFs where the text layer is already sufficient. It can also introduce noisier OCR text when the native text should be preferred.

I would like an opt-in OCR option that skips OCR for pages where Docling has already parsed a usable text layer, while preserving OCR for scanned or sparse-text pages.

Proposed API:

```python
OcrOptions(skip_text_layer_pages=True)
```

Expected behavior:

- default behavior remains unchanged
- `force_full_page_ocr=True` still takes precedence
- pages with enough native text cells return no OCR rectangles
- sparse text, such as a footer/header on a scanned page, does not suppress OCR
- existing bitmap coverage behavior remains unchanged for pages without a usable text layer

This should live in Docling's OCR rectangle selection path so it works for all OCR engines and all callers.

Related context: #2029 and #2036.

### Alternatives

A caller can preflight PDFs outside Docling and set `do_ocr=False` for the whole document. That works for some applications, but it is coarse: one mixed PDF can contain both born-digital pages and scanned pages.

Another option is `do_ocr="auto"`, but that changes the type and meaning of an existing boolean field and overlaps with `ocr_preset="auto"`, which already refers to OCR engine selection.