* Threading lock structure is very slow if it becomes starved of CPUs
* Fix "!!!" points
* Rationalise naming further (e.g. use of "lock" for both peak phases
  and resynchronisation at transients)
* LADSPA plugin has too much "artificial latency"
* LADSPA plugin could do with switches for transient/phaselock
* LADSPA plugin probably doesn't want to go any higher than about +2 octaves
* Vamp plugin argument no longer properly reflect internal parameters
* Solve "for very long stretches" issue in calculateSizes()
* Sort out resynthesis gain

Rejig crispness options:

0 - no transients, no peak lock, long window
1 - no transients, no peak lock, normal window
2 - no transients, peak lock, normal window
3 - band-limited transients, peak lock, normal window
4 - transients, no peak lock, normal window  * - is this a good option to have?
5 - transients, peak lock, normal window
6 - transients, no peak lock, short window