Expand CUDA tape support

[erick-xanadu]

1. Add support for kwargs
2. static_args : This is a bit funny since we are interpreting the function and not compiling.
3. abstracted_axes
4. Test and add support for other devices.
5. control flow operations.
6. quantum control of arbitrary gates.
7. adjoint operations should create their own parameters.
8. Determine version compatibility.
9. Fix samples
10. Fix state
11. Better error messages for these four use cases:
    1. expval(float * obs) (get vague JAX error)
    2. expval(obs + obs) (get vague CUDA quantum error)
    3. gradients (get vague JAX error)
    4. counts/sample without shots specified (get a vague JAX error)
12. AOT compiling
13. Caching