codegen: implement real Eigen matrix-multiply (@) support

[petercorke]

Summary

bdsim.codegen has no C++ implementation for Python's @ (matrix multiply) operator -- CppEmitter.expr_BinaryOp's @ case fails loudly at generation time ("matrix multiply (@) has no C++ implementation yet") rather than emitting a call to a matmul() helper that was never actually defined (the previous behavior: silently generated C++ that failed at compile/link time instead of generation time).

This is independent of continuous-time block support (bdsim#93, closed wontfix) -- that issue is specifically about continuous blocks not being pursued at all; this one is about the @ operator itself, which is on the roadmap regardless. @ shows up in sampled/stateless blocks too -- e.g. PROD's matrix branch (prod = prod @ input), reachable today for any PROD block fed matrix inputs.

Proposed fix

Eigen's own Matrix::operator* already performs real matrix multiplication (not elementwise) for two Matrix-typed operands, so a first cut is likely cheap:

// Python: prod @ input  ->
(prod * input)   // when both operands are Eigen::Matrix-typed

Also needed for PROD's / (matrix division -> np.linalg.inv, which maps to Eigen's .inverse() for square matrices) and, if continuous-block support is ever revisited, LTI_SISO-style self.C @ x.

Scope note

Not the same bug as bdsim#93 -- please don't close this alongside that one; they're unrelated except for sharing a symptom.

[petercorke]

Correction/addition to the summary above: matrix multiply (@) has no C++ implementation at all -- this isn't specific to PROD's matrix branch or to continuous-time blocks. Confirmed it also hits LTI_SS_S/LTI_SISO_S today: unlike LTI_SISO (continuous, out of scope via #93), these are sampled/clocked -- genuinely in scope for this codegen effort -- but their output()/next() do real self.C @ x/self.A @ x + self.B @ u state-space math, so they hit this same failure unconditionally, nothing to do with #93. Likely other blocks in the library hit this too; hasn't been exhaustively audited. Treat this as a general operator gap, not a small, contained list.