Read logic must consider unicode character lengths

[jaredpar]

The SourceTextStream type was operating under the assumption that
Encoder.Convert was a non-throwing method so long as it was passed a
destination buffer with at least one byte available for writing. The
actual contract for Convert is it will not throw so long as it is able
to write the result of converting at least one character to the
destination buffer (or there is nothing to convert). In that case it
will throw an ArgumentException indicating it attempting to do work but
was unable to do so.

The SourceTextStream type processes the characters in chunks according
to the count passed into Read. This caused a bug when a character which
was represented with more than one byte value was at the end of a
logical chunk of text. The Converter would convert all the chars except
the last one. But SourceTextStream continued processing because there
was at least one byte left in the destination buffer and hence an
exception was thrown.

The fix is to not check for count > 0 when processing but instead count
>= the maximum number of bytes the encoding could produce for a single
character.

Note: I did consider calling GetByteCount here instead but decided
against it. It essentially forces the encoder to do the work of
decoding the lead byte twice on every iteration of the loop. Seemed
better to keep the simple worst case check here.

closes #1197

[jaredpar]

CC @pharring, @gafter , @VSadov, @agocke

[gafter]

👍 This is classic!

[agocke]

LGTM

[VSadov]

LGTM