core: add processor buffersize (default 65536), required to be a power of 2 and ...

core: add processor buffersize (default 65536), required to be a power of 2 and >= 4096
for proper support of fft-based analyzers and similar
refactor api:
- improve documentation
- remove IEncoder.finish(), same as process() with nframes < buffersize
- remove format(), description() and such from IDecoder, a decoder can handle multiple
  formats so this information is dynamic
- add duration(), samplerate(), channels() etc... to IDecoder, obsoleting all
  corresponding analyzers
- add IGrapher.process()
- require IGrapher.render() to return a PIL image instead of streaming binary data,
  and move the width/height arguments into the constructor
- add IAnalyzer.process()
- add IValueAnalyzer and let dc, maxlevel and meanlevel analyzers implement it
- move result() into IValueAnalyzer. This method may not be adequate for other types of analyzers