Deprecate bounds in parameters

[walterxie]

All parameters inherited from Tensor should not have the property: lower and upper bounds. Their allowable value range is instead determined by the sampling method (e.g., prior distributions) and by their domain.

The lower and upper fields and setter methods in all parameter classes are deprecated and may be removed in the future if there is no justification for their existence.

[NicoNeureiter]

I like the idea of retiring the lower and upper inputs, but some operators are using the the corresponding attributes for early direct rejects, which might be relevant for performance (also pointed out by @alexeid in #7).

Do you think it would make sense to keep the lower and upper attributes (not the inputs) and set them dynamically via the prior distribution (e.g. in TruncatedRealDistribution or Uniform)?