[BUG] `XGBoostLSS` hardcodes stabilization, response and loss params

[marrov]

Describe the bug
Current implementation of XGBoostLSS seem to hardcode and override some parameters in _fit () before passing it to the original implementation, which does not respect the user provided inputs in the init.

xgblss = XGBoostLSS(
    xgblss_distr(
        stabilization="None",  # Should be self.stabilization
        response_fn="exp",      # Should be self.response_fn
        loss_fn="nll",          # Should be self.loss_fn
    )
)

Am I reading this wrong or was this done intentionally?

[fkiraly]

looks like a bug that the reviewer (me) did not spot