Improve FunctionOperator inverse error

[AshtonSBradley]

Summary

Closes #251.

This PR improves the error users see when applying inverse operations to a FunctionOperator constructed without op_inverse. Instead of falling through to a Nothing callable MethodError, the inverse paths now throw an ArgumentError that tells users to provide op_inverse when constructing the operator.

Root Cause

has_ldiv already reported that a FunctionOperator without op_inverse cannot be divided by, but the \ and ldiv! implementations called L.op_inverse directly. When op_inverse === nothing, that produced the opaque objects of type Nothing are not callable failure from issue #251.

Changes

• Add an internal _assert_has_inverse guard for FunctionOperator inverse operations.
• Apply the guard to both out-of-place and in-place inverse paths: F \ v, ldiv!(w, F, v), and ldiv!(F, v).
• Add regression tests confirming missing inverses throw ArgumentError and mention op_inverse.

Type Stability / Allocations

The success path keeps the existing public API and inverse dispatch shape. The added guard is a small predicate check before calling the already-stored inverse function; the error message allocation only occurs on the missing-inverse error path.

Validation

• julia --project=. -e 'using Pkg; Pkg.test(; test_args=["func.jl"])' passed with 823 passing tests and 2 broken tests.