Version 0.0.4 released.

• Add support of ISO-8859-1 and ISO-8859-15 for French.
• Re-enable Hungarian language models (ISO-8859-2 and Windows-1250) which used to conflict with other charsets (should be better now).
• Differentiate ASCII detection and detection failure.
• Improve single-byte charset detection confidence algorithm (fixes for instance Windows-1251 Russian text detection).
• "UTF-16" is now outputted with endianness information (UTF-16LE/BE).
• Add UTF-32 BOM detection.
• Discard single byte charsets upon illegal codepoint detection.
• Internal redesign of single-byte charmaps with more semantics, and variable sample size length (different languages have different sizes of grapheme lists).
• A lot more test files (33 successful unit tests should be successful with make test).
• Adding python scripts to generate language models from Wikipedia data in a single command.