This is not used by any of the espeak-ng voices and languages.
Additionally, this functionality would be superceded by support
for specifying the language used by different scripts in the
language argument on the command line.

1.  Store the encoding enumeration values in the Translation
    object, instead of the charset table.

2.  Use the encoding.c charset table data instead of the ones
    in translate.c.

3.  Remove the charset language file option -- it is only used
    in the Arabic language file, but is used incorrectly there.

4.  Specify ISO 8859-6 for the 8-bit encoding for Arabic instead
    of UTF-8, so that espeakCHARS_8BIT and espeakCHARS_AUTO work
    correctly for Arabic.

This occurs in ReadVoiceFile (voices.c) when calling isspace to
isolate the attribute name. It happens because the msvc
implementation of isspace asserts that the character is in a valid
range, which fails for non-ASCII characters.

This commit resolves the problem by calling iswspace, which uses
the ucd-tools implementation.

This is to support splitting language specification (dictionaries)
from voice specification (phon*).

With the original eSpeak Windows port, the MBROLA voices had to
be present in an mbrola folder of the espeak-data. This will be
different in espeak-ng, allowing the MBROLA voice files to be
located elsewhere on the system.

This directory is no longer available, so the check is now
redundant.

Moved include/espeak/speak_lib.h to include/espeak-ng/speak_lib.h and
created a symlink from the original location.

These heuristics were designed to guess the location of the voice
files without searching the voice list. The heuristics are very
implementation dependent, and easily break when voices move around.
The voices are still found correctly, so these heuristics are
redundant.