Found while building an aircraft trim example using LUT1D as a CL(alpha)/
Cm(alpha) lookup table. LUT1D's underlying scipy interp1d call, for a
single scalar output, returns a 0-d numpy array; Register.update_from_array
did len(arr) on anything isinstance(arr, np.ndarray), which raises
TypeError for 0-d arrays. Special-case 0-d arrays as scalars.

Root cause of why this went uncaught: tests/pathsim/blocks/test_table.py
had two classes both named TestLUT1D. Python silently let the second
shadow the first at module import, so pytest never collected the first
class's tests (test_embedding_siso, test_update_siso, etc.) -- including
the coverage that would have caught this. Renamed the first class to
TestLUT1DEmbedding to restore it.

Unshadowing surfaced two more latent issues, fixed here:
- Register.__setitem__ didn't tolerate a 1-element numpy array assigned to
  a scalar int index (numpy 2.x: "setting an array element with a
  sequence") -- extended the existing 0-d scalar-coercion branch.
- test_linearization_siso asserted the wrong expected value: linear
  interpolation between grid points (1, 1) and (2, 4) at x=1.5 correctly
  gives 2.5, not the exact quadratic 1.5**2=2.25 the test asserted.

test_sensitivity in the same file imports pathsim.optim.value.Value, which
does not exist anywhere in the current src/ tree -- looks like a stale test
for removed/never-merged AD functionality. Marked @unittest.skip with an
explicit reason instead of guessing at reimplementing it; worth reporting
separately.