Summary

The README's stability-check snippet errors when run as written:

```python
A = model.recurrent.injection.get_A()
rho = torch.linalg.eigvals(A).abs().max().item()

RuntimeError: linalg.eig: The input tensor A must have at least 2 dimensions.

```

`LTIInjection.get_A()` returns a 1-D tensor of shape `(dim,)` — the diagonal of a diagonal state matrix — so `torch.linalg.eigvals` rejects it. The eigenvalues of a diagonal matrix are simply its diagonal entries, so the correct (and simpler) form is `A.abs().max()`.

Reproduced on a fresh install (`open-mythos==0.5.0`, Python 3.11, PyTorch 2.8) using the exact README example. After the fix the example runs end-to-end and prints `ρ(A) = 0.3679` (= `exp(-1)` at init, matching the `log_dt = log_A = 0` parameterization).

Test plan

• Ran the README snippet on a clean install — errored before, prints `ρ(A) = 0.3679` after
• Cross-checked against `torch.linalg.eigvals(torch.diag(A)).abs().max()` — same value
• Verified `ρ < 1` holds at init by construction (`A_i = exp(-exp(log_dt + log_A))`)