Update to HiddenMarkovModels v0.6

[gdalle]

I recently released a breaking version of HiddenMarkovModels, so here's a small PR updating the compat bound and adding a few views in passing for speed

[dmetivie]

Thanks! I just up version and added your new AbstractVectorOrNTuple type for seq_ends.

[gdalle]

Thanks! I just up version and added your new AbstractVectorOrNTuple type for seq_ends.

This is not documented so it is not part of the public API and could break in patch release. At the moment you should not rely on this functionality.

[gdalle]

The main reason I introduced it was to activate or deactivate multithreading based on dispatch, so that a single sequence (passed as seq_ends=(T,)) would not trigger the overhead of multithreading. It is mostly an internal change.

[dmetivie]

Do you recommend removing the type annotation for seq_ends in StatsAPI.fit! function, then?
Keeping the old AbstractVector{Int} was making the code error.

[gdalle]

I would recommend simply removing the annotation on seq_ends in your custom fit!, since now it must be able to accept both a vector and a tuple. I believe that's how I do it in the new docs but I'll gladly take suggestions to document this a bit better.

[gdalle]

And lastly, you don't need to release a breaking version when you update a dependency like that, except when it has an impact on your package's own user-facing API. So here tagging v0.2.1 would probably have sufficed. But we can discuss it face to face in Toulouse ;)

[dmetivie]

Dam! I can always revert the pkg number since it not release yet?

Can't wait (I'll be there only Tuesday, unfortunately)!