Fix LIF row padding with mixed memory blocks

[cgohlke]

This PR improves plane reading robustness especially for LIF files where non-image memory blocks (for example, timestamps) are interleaved with image data:

• Prefer explicit per-row stride from BytesInc metadata when available, instead of inferring row padding from adjacent memory block offsets.
• Keep inferred-gap logic only as a legacy fallback when row stride metadata is unavailable.
• Restrict fallback padding inference to rows that are not naturally 4-byte aligned (based on row bytes, not pixel width).
• Clamp inferred fallback padding to non-negative values.

See https://forum.image.sc/t/problems-reading-some-widefield-leica-lif-files-with-bioformats-in-fiji/121584 for the original bug report and a test file.

[cgohlke]

Note that I have changed the fallback condition from getSizeX() % 4 to (getSizeX() * bpp) % 4. I didn't quite understand the original intent, neither am I sure if the fallback is still necessary...