Fix continuation-byte check for UTF-8 chars > 2 bytes.

simplify utf-8 continuation byte checking, using a loop in stead of separater 2/3/4-byte versions.

src/merge3.c: In function ‘merge_3way’:
src/merge3.c:445:9: warning: ‘zPivot’ may be used uninitialized in this function
src/merge3.c:446:9: warning: ‘zOrig’ may be used uninitialized in this function
src/merge3.c:447:9: warning: ‘zOther’ may be used uninitialized in this function

Don't try to convert binary files.

Factor out main part of UTF-8 check to macro

The earlier check for FEFF was not sensible: A BOM in the middle of an UTF-16 file could easily arise when two UTF-16LE or twe UTF-16BE files are concatenated. Nothing goes wrong in that case, the BOM should just be displayed a zero-width-space. Therefore, diff should not bail out.

let looks_like_text() give different values for UTF-16 BE/LE. Not used yet.

For UTF-16, the line-length limit is set to 2/3th of the line length limit for other text, because UTF-16 -> UTF-8 conversion can increase the line length (in bytes) by max 50%. This guarantees that a UTF-16 diff can be made by converting the two UTF-16 files to UTF-8 and then do a normal diff.

re-use looks_like_blob

Trying to commit a file with an UTF-16 BOM will now give a warning, just as a file containing crlf

The --args command line option now accept UTF-16 files as well