Previously vte used to convert the whole chunk of input from
the PTY into UTF-32 in one go, and feed that to the parser.
Replace this with a streaming converter, which consumes the
bytes one-by-one and feeds the result to the parser whenever
a unicode character falls out of it.

Also simplify the chunk handling by using C++ standard classes
to manage the queue (and stack of reserve chunks) instead
of a hand-grown list handling code.

As a result of the converter change, vte currently doesn't work
with non-UTF-8 I/O encoding. This will either be reimplemented
or non-UTF-8 support will be dropped completely (to be discussed
in bugzilla).

The change fixes a bug in parser-cat (and potentially also in vte
itself, haven't been able to check) where occasionally a
character at the end (or start?) of a chunk was discarded. That can
be observed by comparing the output of 'parser-cat --charset "UTF-8"'
(old converter) vs just parser-cat (new converter).