feature request: support for (matrix)-normal-inverse-wishart

[murphyk]

It would be great if you could implement the (M)NIW code we wrote at https://github.com/probml/ssm-jax/blob/main/ssm_jax/distributions.py#L107 as part of core tfd, so we can avoid issues such as https://github.com/tensorflow/probability/issues/1617 in the future.

Conjugate Bayesian updating of the parameters (as in https://github.com/probml/ssm-jax/blob/main/ssm_jax/distributions.py#L282) would also be nice :)