Issue 183: buffering of latent delays

[seabbs]

This PR closes #183 by adding support for buffering in all currently supported observation models and in LatentDelay.

I am quite unhappy about the current approach because:

• Requires support in all observation models that will be used with LatentDelay
• Has different behaviour when completely missing a vector of missingness is passed in

All that being said I think we should review this as a fix and use any major review comments in a new issue to improve this. One option we could consider is no longer support passing just y_t = missing and instead enforce that this must be a empty vector (and helping people achieve this). Just this change would streamline at least part of this.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.16%. Comparing base (1e9390b) to head (5554f74).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #250      +/-   ##
==========================================
+ Coverage   93.07%   93.16%   +0.08%     
==========================================
  Files          43       43              
  Lines         419      424       +5     
==========================================
+ Hits          390      395       +5     
  Misses         29       29              

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

[SamuelBrand1]

This looks good.

As a note: this PR implies redoing the underlying data generating for the pipeline (which is good and essentially what we are fixing here).