timesteps instead of number of timepoints

Use time steps (in dataframes) instead of numpy arrays. index argument that can pass a dataframe (inferred or explicitly specified which would override) or a numpy array with index (if missing, inferred).

Update:

Weights function
timecorr (format data, not hypertools) force to return timecrystal
helpers:
- all weights functions (should take index instead of number of timepoints)
- wcorr (takes weights matrix now, but make sure that the timepoints are processed in the correct order)
- smooth function (really nonfunctional after... same with timepoint decoder... and predict)