Add scalar dispatch for SymbolicNeuralNetwork (fixes #83)

[ChrisRackauckas-Claude]

Summary

This PR adds a Number dispatch to StatelessApplyWrapper that allows using scalar variables directly without wrapping them in arrays, as suggested by @ChrisRackauckas in #83.

Previously, users had to write:

sym_nn([X], θ)[1]

Now they can write:

sym_nn(X, θ)[1]

The array syntax still works for backward compatibility.

Implementation

Added a new method to StatelessApplyWrapper at src/ModelingToolkitNeuralNets.jl:124-126:

function (wrapper::StatelessApplyWrapper)(input::Number, nn_p::AbstractVector)
    wrapper([input], nn_p)
end

This method wraps scalar inputs in an array and delegates to the existing AbstractArray dispatch.

Test plan

• Added comprehensive test in test/scalar_dispatch.jl demonstrating scalar dispatch usage
• Verified backward compatibility with existing array syntax
• Ran full test suite with Pkg.test() - all tests pass (27/27)
• Formatted code with JuliaFormatter using SciMLStyle

Fixes #83

🤖 Generated with Claude Code