v19.0.0

Bumped the major version for a "breaking" change - but unlikely to impact any users:

• NumericalInverse now uses implicit differentiation instead of erroring. This impacts for example block neural autoregressive flows. Only "breaking" change is if users were to e.g. expect and catch the error which no longer occurs.

What's Changed

• Auto-differentiable numerical inverse & efficient materialization of BNAF masks by @noahewolfe in #234
• Remove unused wrapper and make flag kw only by @danielward27 in #235

New Contributors

• @noahewolfe made their first contribution in #234. Thanks!

Full Changelog: v18.0.0...v19.0.0