Use the time axis of transposed inputs in discrete-time forced_response - #1255

Fixes #931.

In discrete time, forced_response() builds the time vector from U when T is not given, but it took the number of steps from U.shape[1] even with transpose=True, where time is along the first axis:

sys = ct.rss(3, 2, 2, dt=0.1)
U = np.random.randn(2, 10)
ct.forced_response(sys, U=U.T, transpose=True)
# ValueError: Parameter `U`: Wrong shape (rows, columns): (2, 10). Expected: (2, 2).

For a single input, U of shape (10, 1) gave a one-step time vector, a "divide by zero" warning in the time step check, and then a shape error. The number of steps now comes from U.shape[0] when transpose=True.

The same branch also expands a scalar U to (ninputs, len(T)) when T is given. With transpose=True that array is transposed again when its shape is checked, so ct.forced_response(sys, T=T, U=1.0, transpose=True) failed with a shape error for discrete-time systems (it works in continuous time). The expanded array is now transposed first in that case.

Added test_forced_response_transpose_discrete in control/tests/timeresp_test.py (one and two inputs, comparing against the non-transposed call). It fails on main and passes here. timeresp_test.py, discrete_test.py, matlab_test.py, trdata_test.py and docstrings_test.py pass locally (486 passed, 52 skipped), and ruff check is clean.