feat: add SemilinearODEFunction and SemilinearODEProblem

[AayushSabharwal]

Generates

f1 = B * x2 + C
f2 = A * x

Where x2 is the upper triangle of x * x' as a linear vector with unnecessary elements zeroed. This is what Symbolics.semiquadratic_form returns.

In terms of codegen, the IIP form uses a DiffCache to avoid allocating a new buffer for x2 each time (and to allow passing duals) and generates the equivalent of

x2 = view(PreallocationTools.get_tmp(...), ...)
x2[1] = ...
x2[2] = ...
...
out[1] = C[1]
out[2] = C[2]
...
mul!(C, B, x2, true, true)

The jacobian for f1 is B * x2' + C' where x2' is the jacobian of x2 and C' that of C. This again uses the DiffCache and generates code with the same structure as above. For sparse forms, it is basically identical except the array from the DiffCache is used as the nzvals of a SparseMatrixCSC.

Still needs a little more implementation for the sparse form, tests and docs.

[AayushSabharwal]

No point running CI right now if no tests or docs are changed.