Inconsistent unit validation error handling for y_unit in diffusion models #248

Description

Unit validation error handling in diffusion_model_base is inconsistent between x_unit and y_unit.

@henrikjacobsenfys isolated the source: When an invalid unit (such as an integer 123) is passed as x_unit, diffusion_model_base explicitly checks that it is convertible by calling _assert_convertible_to_mev, which correctly raises a scipp.UnitError. However, passing an invalid unit to y_unit bypasses explicit validation and fails later with a generic TypeError.

This inconsistency forces unit tests to expect different exceptions for the exact same logical error. This was highlighted in #246 , but appears to be a general issue. Multiple unit tests in tests/unit/easydynamics/sample_model/diffusion_model currently use the same hotfix where invalid y_unit inputs are explicitly parameterized to expect a TypeError.

Steps to Reproduce

In the unit test (e.g., inside test_chudley_elliott_diffusion.py), parameterizing invalid unit tests currently requires expecting different exceptions:

# For x_unit, explicit validation correctly raises a scipp.UnitError
(
    {'x_unit': 123, 'scale': 1.0, 'residence_time': 1.0, 'jump_length': 1.0},
    UnitError,
    'Invalid unit',
),
# For y_unit, we currently have to expect a TypeError as a workaround
(
    {'y_unit': 123, 'scale': 1.0, 'residence_time': 1.0, 'jump_length': 1.0},
    TypeError,
    None,
),

Behavior

• Actual Behavior: Invalid x_unit raises UnitError, while an invalid y_unit falls through to raise a scipp.TypeError. Unit tests are currently hardcoded to accommodate this mismatch.
• Expected Behavior: Both invalid x_unit and y_unit should be caught during initialization and raise a UnitError with a clear error message (e.g., 'Invalid unit').