Logisets cannot be used with the standard Julia data interface (`MLUtils.numobs`/`getobs`)

Julia's ML stack - MLJ, Flux, DataLoaders and the batching/splitting utilities around them -
interoperates through `MLUtils.jl`'s `numobs`/`getobs` interface. A container implementing it works
with `splitobs`, `DataLoader`, shuffling and batching for free.

`SoleData` does not reference `MLUtils` anywhere: it is absent from `Project.toml` (deps, weakdeps and
extensions), and a code search across the repository returns zero matches. Slicing is instead offered
through the package's own `slicedataset`.

The consequence is that a logiset cannot be handed to standard Julia ML tooling without the user
writing an adapter - a barrier at exactly the moment someone is deciding whether to adopt the
ecosystem.

`DataFrames` and `Tables` are already dependencies, so the interoperability instinct is clearly there;
this is the one connector missing from it.

### What would close it

`MLUtils.numobs` and `MLUtils.getobs` for `AbstractLogiset`, most naturally as a package extension so
it costs nothing when `MLUtils` is absent - the same pattern `SoleModels` already uses for
`DecisionTree` and `XGBoost`.

Reported after an outside survey of the ecosystem's interoperability surface. If logisets deliberately
do not fit that interface, that is worth stating in the docs and this issue can close.