The PDF specification explicitly says that after "a" to "b", the numbering follows "aa", "bb", ..., "zz" and not "ab", "ab", "az", "ba",...
59d6599 should be reverted

We should support two different parameters, the HTML "href" value, which refers to the link target, and independently the text that is visible to the user.

Also document that `Xml.add_paragraph()` isa context manager.

The link target must always be specified. Only the its display in the text is optional.

Additional safeguard when field is being deleted via `page.delete_widget()`:
In this case, the underlying annotation may not have been re-established (attribute "_annot"), which will there for be checked.

Python C function `Py_BuildValue("s", fontname)` will fail if fontname is not UTF8-encoded.
Use PyUnicodeRawEscape function for fontnames instead - like everywhere else in PyMuPDF.

helper-annot.i (JM_get_annot_xref_list):
Only include actual annotations in list - no longer "null" objects, etc.
Refers to #2063

helper-python.i (class linkDest):
Treat "&zoom=nan" string in `uri` like "&zoom=0".

helper-stext.i (various):
* Replace font ascender / descender returns with sane values when necessary.

* JM_char_quad:
Guard against rectangles transformed from quads that are not completely located in the positive quadrant.
This may happen with incorrectly interpretated Type 3 fonts.

fitz.i (Document.extract_image):
If multiple filters for image streams are present, function `pdf_load_stream()` must be used instead of `pdf_load_raw_stream()` to determine the image type.

For correctly detecting a rectangle,  the equality of x- and y-coordinates of the respective 4 corners must be confirmed.
We did not check x-coordinates.