[Contribution] Optional Lasso penalties on regressor and trend terms.

Hi

Thanks for helping to maintain this amazing library!

Recently I did some work where I added a Lasso penalty to the regressor, trend and offset terms (i.e. enable the option to change the prior on parameters `k, m, delta` and `beta` to be `double_exponential`, with scale parameter set by the user).

This had a few benefits:

- If a regressor is deemed not useful by the model, it is shrunk close to zero
- If the trend and the regressors interact (e.g. if you have a very strong correlation between trend and a regressor), the trend can do some strange stuff unless it's set to 'flat'. The Lasso penalty can help to avoid this issue automatically.

I was wondering if there would be any interest in adding this as a contribution?

Completely understand it's a big change so you may not want to add it to core. I just thought it was worth asking beforehand, especially as then I could check about best design practices for this.