Don't fail if EndOfInformation is omitted

[jasinb]

Some real-world LZW-compressed TIFFs omit the trailing EndOfInformation code (libtiff accepts this with a warning). Don't fail with UnexpectedEof in this case.

[jasinb]

I was unsure if the correct error message should still be UnexpectedEof if we get NoProgress even with some input remaining, but I opted to keep the diff minimal. Another choice would be InvalidData or something.

[jasinb]

The check failures seem unrelated.

[197g]

This was likely caused by an issue in the lzw decoder that should have been fixed by #392. Do you have any test images to verify? At least I'm assuming that we do want to continue detecting streams that do not contain all necessary samples.