Optimize wave-form computation
It seems that we have room for improvement for wave-form computation. Haven't looked at the code lately, but some random suggestions:
- The waveform needs to be shared across clips from the same bin. Cutting should essentially be a no-op for both sides of the clip
- We could conform the audio (computing an uncompressed file at the correct frame-rate). That could potentially help for precise timings too, but the downside is that it requires big cache files.
Additional remark: some reports seem to indicate that it gets worse when the zoom level is high, we need to investigate.