Update the SDM type with a trained flag

Can be done without breaking backwards compat

1) add a `trained::Bool = false` field
2) `train!` sets `trained` to `true`
3) `hyperparameters!` sets `trained` to `false`
4) `predict` throws an `UntrainedModelError` when called on a model with `!trained`
4) the `show` method started with `[UNTRAINED]`