DocumentToImageContent still fails the whole batch with KeyError when a PDF page can't be converted #12911

Describe the bug
Since #12872, DocumentToImageContent returns None for an invalid document instead of failing the whole batch. This still doesn't work when a document points to a PDF page that can't be converted: the page is past the end of the PDF, or the PDF can't be read. _convert_pdf_to_images skips those pages and logs a warning, but _batch_convert_pdf_pages_to_images then looks them up with page_number_to_image[page_number] and raises a KeyError. The valid documents in the batch are lost too, including when it runs inside LLMDocumentContentExtractor.

Error message

KeyError: 99
  File "haystack/components/converters/image/image_utils.py", line 337, in _batch_convert_pdf_pages_to_images

Expected behavior
The document gets None in image_contents and the others are still converted. That's what the run() docstring describes and what test_run_none_images expects. That test only passes because it mocks _batch_convert_pdf_pages_to_images to return {}.

Additional context
In practice this happens when stored documents point to a PDF that was later replaced with a shorter version, when a PDF is corrupted or only partly downloaded, or when a splitter assigns a page number past the end (see #12909).

To Reproduce

from haystack import Document
from haystack.components.converters.image import DocumentToImageContent

pdf = "test/test_files/pdf/sample_pdf_1.pdf"
DocumentToImageContent().run(documents=[
    Document(content="", meta={"file_path": pdf, "page_number": 1}),
    Document(content="", meta={"file_path": pdf, "page_number": 99}),
])
# KeyError: 99

An unreadable PDF (for example a file containing just %PDF-1.4 not a real PDF) raises KeyError: 1 the same way.

I'd like to work on a fix.

System:

• OS: Windows 11
• Haystack version: main (3.3.0-rc0)