Reparametrize raisedcosine

[BalzaniEdoardo]

This edits introduce a one extra hyper-parameters for initializing the linear raised cosine bumps, and an two one for the log-raised cosines:

• width: this hyper-parameter is set for both linear- and log-spaced raised cosines. It controls the width of the cosine bumps, default is 2. The original implmentation was assuming this to be 1, but that produces aliasing
• time_scaling: this hyper-parameter is available only for log-spaced raised cosines and controls the logarithmic stretch magnitude. as time_scaling approaches to 0, the transformation becomes linear. large time_scaling result in basis that are tightly concentrated around t=0
• enforce_decay_to_zero: boolean, if true, it enforce the last basis to decay to zero.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (2dcf1b2) 94.88% compared to head (664169d) 95.04%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #50      +/-   ##
==========================================
+ Coverage   94.88%   95.04%   +0.16%     
==========================================
  Files          10       10              
  Lines         840      868      +28     
==========================================
+ Hits          797      825      +28     
  Misses         43       43              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[billbrod]

This looks pretty much good to me! Some small changes scattered throughout, and then the following things I think we should discuss:

• Basis docstrings. I think they should have a line or two before parameters that describe what the basis looks like and when you would use them. E.g., do they tile uniformly, are they each about the same size, what types of variables make sense (visual input, head direction, etc.). And then the footnote to references belongs in those lines (it looks weird in the docstring in jupyter or whatever to have the link right at the top). This is probably out-of-scope for this PR, in which case we should open an issue for it.
• We need an explanation of the particularities of the different parameters for the Raised Cosine parameters. Much of that will end up in our separate repo, but we need some here. Do we want to hold off on that until we've worked through it in the separate repo and then add something here? (in which case, we should open an issue)

[BalzaniEdoardo]

I changed the code so that the peak location is computed so that the decay to zero of the last basis element happens as late as possible. this is not equivalent to what we had before, (it would be if i use np.ceil(width) instead of width to compute the last peak location.

[billbrod]

Suggested change

	# - as the time_scaling tends to inf, basis will be small and dense around 0 and
	# progressively larger and less dense towards 1.
	# - as the time_scaling tends to inf, basis will be small and dense around 0 and
	# progressively larger and less dense towards 1.

[billbrod]

I think this looks good to me!