GeezerVison

Closes #1565.

Problem

Package._gather_image_parts walks the package relationships and appends the target of every image relationship to the ImageParts collection, casting it to ImagePart:

def _gather_image_parts(self):
    for rel in self.iter_rels():
        if rel.is_external:
            continue
        if rel.reltype != RT.IMAGE:
            continue
        if rel.target_part in self.image_parts:
            continue
        self.image_parts.append(cast("ImagePart", rel.target_part))

The cast is only a type hint. If an image uses a type that does not load as an ImagePart (an SVG, or an image with broken or non-standard EXIF data as in the report), the target part is a plain Part, which has no sha1.

Once such a part is in the collection, the next add_picture call runs ImageParts._get_by_sha1, which reads .sha1 on every collected part, so it raised 'Part' object has no attribute 'sha1'.

Fix

Skip parts that have no sha1 while matching:

for image_part in self._image_parts:
    if not hasattr(image_part, "sha1"):
        continue
    if image_part.sha1 == sha1:
        return image_part

This mirrors the same fix already made in python-pptx (_ImageParts._find_by_sha1, commit 4c0fd125), which the issue links to.

Tests

Added it_skips_parts_without_a_sha1_when_matching in tests/test_package.py. It puts a plain Part (no sha1) ahead of the matching ImagePart in the collection and asserts _get_by_sha1 returns the ImagePart rather than raising. The test fails on the current code with the reported AttributeError and passes with the fix.

Verified with pytest tests/test_package.py (9 passed) and ruff check on the changed files. The unrelated collection failures elsewhere in the suite on my machine come from a pyparsing version mismatch in the cxml test helper and are present on a clean checkout too.