Modular OutputBuilder + better demo performance

[juanmc2005]

Changelog

• Add RTTMWriter observer to write system outputs
• Add accumulate_output() and buffer_output() operators to concatenate system outputs either by accumulating all past predictions or by keeping only the last ones
• Add RealTimePlot observer to draw system outputs in real time
• OnlineSpeakerDiarization pipeline yields None waveform if output_waveform=False for consistency
• Remove redundant sample rate parameter in FileAudioSource
• Make accumulate_output() more efficient by pre-allocating the audio buffer and hence reducing the number of costly concatenate operations