Avoid signed-shift UB in pngfix get32

[soul-sol]

Fixes #779.

get32 promoted each input byte to signed int before shifting. For a high byte such as 0x98, the expression 152 << 24 is not representable as int and triggers undefined behavior.

Reuse the existing PNG_U32 helper so every byte is converted to png_uint_32 before the shift. This keeps the decoded value unchanged while making the arithmetic defined.

Validation:

• Reproduced the attached issue input on current libpng18 with Clang UBSan before the patch: runtime error: left shift of 152 by 24 places cannot be represented in type int.
• The same input produces no UBSan diagnostic after the patch.
• Full UBSan-enabled CTest suite: 37/37 passed.
• Strict C90 Clang syntax check with warnings as errors passed.
• git diff --check passed.

Signed-off-by: solim 7576268+soul-sol@users.noreply.github.com

[jbowler]

@soul-sol where does "PNG_U32" come from?