- introduce parameter gp_type
 - remove parameter method
 - validate parameter consistency
 - compute Lp separatly
 - more testing

The key changes are:

- **Old Behavior**: `sigma` explained noise on the input data, impacting
  the conditional mean functions and thus the prediction. This
  prediction reflected the mean over all functions and all possible
  inputs given the variability of the input data indicated by `sigma`.
- **Intermediate Behavior**: `sigma` now only inflated the uncertainty
  of the prediction, assuming the input is fixed. The new behavior is
  more sensible for the `DensityEstimator`, where the uncertainty of the
  input is quantified only by ADVI inference but the values `y` are
  already mean values of the GP.
- **Introduction of `y_is_mean` Parameter**: A boolean parameter
  `y_is_mean` is added to the `conditional_mean` that computes the
  predictive function of the GP. If `y_is_mean=True`, the `y` values are
  considered a fixed mean, and `sigma` only reflects the uncertainty
  estimate. If `y_is_mean=False`, the values `y` are treated as a noisy
  measurement, leading to a potentially smoothed value at corresponding
  locations `x`.

This update brings clarity to the treatment of values we condition on in
the Gaussian Process and allows controlling if they are seen as the mean
of a Gaussian Process or noisy measurements. It promotes better
alignment with the underlying statistical principles and the
requirements of the `DensityEstimator` and `FunctionEstimator` ensuring
backwards compatibility.

…r naming