Improve WCPG estimation technique

[hidmic]

Worst case peak gain in ltitop is currently estimated by brute force:

ltitop/src/ltitop/models/analysis.py

Lines 45 to 56 in d75f165

	# TODO(hidmic): do better than brute force
	WCPG = D
	Ap = np.eye(A.shape[0])
	WCPG_terms = np.zeros_like(D)
	for n in range(math.ceil(nmax / 100)):
	WCPG_terms[:] = 0.
	for _ in range(n * 100, (n + 1) * 100):
	WCPG_terms += np.abs(C @ Ap @ B)
	Ap = Ap @ A
	WCPG += WCPG_terms
	if np.all((WCPG_terms / WCPG) <= rel_tol):
	break # early

Accuracy guarantees are thus quite loose, and the resulting performance is awful. It calls for a better approach.

[hidmic]

I've attempted to use fixif.WCPG for this, but performance isn't great either and the requirement on the system matrix A to be diagonalizable is a bit limiting.