fix: fix assert_jac_length_header

[AayushSabharwal]

Close #3554

[AayushSabharwal]

I can add something like the MWE in the issue as a test if we want, just not sure if exact numerical equality is something we (or SU) can guarantee in these cases.

[ChrisRackauckas]

Let's add it as a test but put a comment to the PR and that it might be disabled in the future. Indeed, we cannot really guarantee that because LLVM could do optimizations that make it false, but at least for right now it seems like a good enough test.

[vyudu]

Just so I understand what's the change in the header here that makes it work with the CSE? Is it appending the assert to the end of the function?

[AayushSabharwal]

No. For CSE to work, it needs to get a symbolic expression. When you interpolate expr.body into a LiteralExpr, all CSE sees is a Julia Expr so it can't do anything.

[AayushSabharwal]

SymbolicIndexingInterface test group fails because SciMLBase needs a hotfix