Buffering handling
After fighting against a buffering bug, it is clear that buffering should be improved... Basically, we should have:
- BUFF* keys used as caller's requirements (ie Snowpack wants to compute TSS24 and therefore needs data up to 24 hours before the start date)
- ProcessingProperties objects containing our internal requirements (from filters and resampling)
Internally, we should always use BUFF_BEFORE+proc.buff_before as well as the matching buffer size. Then all the necessary data is always there and there is no unnecessary rebuffering.