Fix the autoencoder

[gdkrmr]

• remove the raw tensorflow v1 code.
• switch to keras3
• api changes require rewriting of tests

[gdkrmr]

see #58

[topepo]

How are things coming along with this?

[gdkrmr]

I don't think I will fix this. I have given it some thought and dimRed should not provide any implementations of method and just wrap other people's packages. Current plan is to just remove the autoencoder and skip the failing test on cran and then resubmit.

Are there any packages on CRAN that provide good implementations of autoencoders?

[topepo]

Are there any packages on CRAN that provide good implementations of autoencoders?

Probably but I've never made the effort to find out 😄

I thought of making one via torch and put it in the brulee package though. I think that torch is easier to maintain in an R package. The keras API changes with no planned deprecation path have been awful. Anyway, a brulle function won't happen this year unless anyone wants to lead that charge (not imply that you do it).

I would remove it it that is what it takes to get a release out.