types: make Image path extension check case-insensitive

[anxkhn]

Image.serialize_model decides whether a string is a file path or already-base64
data. When a string is not an existing file, it checks the extension to tell "a
path that does not exist" apart from raw base64:

if self.value.split('.')[-1] in ('png', 'jpg', 'jpeg', 'webp'):
    raise ValueError(f'File {self.value} does not exist')

That comparison is case-sensitive, but image files routinely carry uppercase
extensions (cameras emit .JPG; case-insensitive filesystems treat .PNG the
same as .png). So a non-existent path with an uppercase extension skips the
File ... does not exist branch and falls through to the base64 handling:

• Image(value='some/path/photo.jpg').model_dump() correctly raises
  File some/path/photo.jpg does not exist.
• Image(value='some/path/photo.JPG').model_dump() instead raises the misleading
  Invalid image data, expected base64 string or path to image file.
• Worse, a bare filename that happens to be valid base64 is silently accepted:
  Image(value='PHOTO.PNG').model_dump() returns the literal string 'PHOTO.PNG',
  which then gets sent in the images array to /api/generate and /api/chat as
  garbage image data instead of failing fast.

The fix lowercases the extracted extension before the membership test so
.PNG/.JPG/.JPEG/.WEBP behave the same as their lowercase counterparts. Non-image
strings (e.g. model_v1.2, plain base64) still fall through to the base64 check
unchanged.

Added a parametrized regression test in tests/test_type_serialization.py
covering the uppercase extensions for both a path-shaped value and a bare
filename; both cases fail on the current code and pass with the fix.

[vaderollama]

LGTM — clean one-liner fix. Case-insensitive extension check prevents uppercase image extensions (.JPG, .PNG) from being misidentified as base64 data. Tests parametrize the common uppercase variants.

@ParthSareen — ready to merge.