* Blockwise optimal linear control ops

* Add jax rewrite to eliminate `BilinearSolveDiscreteLyapunov`

* set `solve_discrete_lyapunov` method default to bilinear

* Appease mypy

* restore method dispatching

* Use `pt.vectorize` on base `solve_discrete_lyapunov` case

* Apply JAX rewrite before canonicalization

* Improve tests

* Remove useless warning filters

* Fix local_blockwise_alloc rewrite

The rewrite was squeezing too many dimensions of the alloced value, when this didn't have dummy expand dims to the left.

* Fix float32 tests

* Test against complex inputs

* Appease ViPy (Vieira-py type checking)

* Remove condition from `TensorLike` import

* Infer dtype from `node.outputs.type.dtype`

* Remove unused mypy ignore

* Don't manually set dtype of output

Revert change to `_solve_discrete_lyapunov`

* Set dtype of Op outputs

---------
Co-authored-by: ricardoV94 <ricardo.vieira1994@gmail.com>