The conversion from UTF-16 to UTF-8 could (until a few days ago) fail
before the end. In that case, the appended nul terminator would not be
converted. A nul terminator is anyway always appended after conversion
and trimming.

If conversion failed on the first character, that nul terminator would
be written at a negative offset, leading to heap buffer "underflow" and
memory corruption. This was fixed but lead to a mismatch in the value of
i_lines depending on the character width.

This change removes the useless pre-conversion nul terminator, and thus
makes trimming work again with single byte character width work. This
fixes reading text files formatted with MS-DOS line endings.