Also allow arbitrary expression dimensionality

Some users with outdated setuptools in their build environments are failing
to build miniKanren from source. A wheel will hopefully arrive shortly for v1.0.5,
solving the problem. For now we can work around it by avoiding v1.0.4.
Note that conda doesn't have this problem, because Conda packages are
pre-compiled.

* Use lapack func instead of `scipy.linalg.cholesky`

* Now skips 2D checks in perform
* Updated the default arguments for `check_finite` to false to match documentation
* Add benchmark test case

* Refactor out _cholesky helper, add empty test

* Remove array and `potrf` copies

* Update test_cholesky_raises_on_nan_input

* Add jax dispatch for CholeskySolve

* Better typehints on user-facing `cho_solve`

* Rename test

* Created some tests that fail due to #1476

* Fixes 1476 and other ways to create a log1mexp

* Reimplemented logmexpm1_to_log1mexp by tracking expm1 and then looking through the clients

* Absorbed the rewrite log1pexp_to_softplus into the new rewrite for log1mexp

* Fixed bug where I forgot to check whether result of is_neg was None or not before proceeding

---------
Co-authored-by: Luca Citi <lciti@ieee.org>

* Avoid pytest warning for variable name

* Respect core type shape in gradient of Blockwise

* Refactor Blockwise L_op

Reuse `cholesky` decomposition with `cho_solve` in graphs with multiple `pt.solve` when `assume_a = "pos"` (#1467)

* Extend decomp+solve rewrite machinery to `assume_a="pos"`

* Update rewrite name in test

* Refactor tests to be nicer

* Respect core op `lower` flag when rewriting to ChoSolve

* Expose `pt.optimize`

* Clean up imports in optimize.py

* Add example notebook for `optimize.root`

* Small updates

* Add doc file

* Link to optimize docs in tensor index

* Add docs to user-facing functions.

* Move `scipy.optimize` imports into `perform` methods

* Use global import strategy

* Remove props and overload __str__

* rerun example notebook

* Implement `optimize.minimize`

* Add more tests

* Code cleanup

* Add RootOp, fix gradient tests (they are failing)

* Correct gradients for `minimize`

* Mypy

* remove debug flag

* minimize works

* Implement minimize_scalar

* Use LRU cache wrapper for hessian option

* Remove useless Blockwise

* Feedback

* use truncated_graph_inputs and refactor

* Implement Root Op

* Factor out shared functions

* Implement `root_scalar`

* mypy 😍

* Add specialized `build_fn` to each Op to handle scipy quirks

* Changes to support float32 inputs

* Check inputs to RootOp

* More mypy

* Fix bug when minimize gets scalar input

Changes:
1. Remove references to c-code which apply to `COp` but not `Op`
2. Fix failing doctests
3. Improve explanation of `make_node`
4. Emphasize distinction between itypes/otypes and make-node
5. Show `L_op` instead of `grad`
6. Show how to test `L_op` and `infer_shape` implementation
7. Simplify explanation of `__props__` and illustrate in example.
8. Introduce more complex multi-output Op to drive these details home
9. Remove old references to numba/ random variable Ops