Fix lumped thermal discretisation with 1+1D current collectors

[chenyi6758-source]

Closes #5878.

"Surface temperature [K]" carries the current-collector domain, so with 1+1D current collectors the lumped cooling term varied in z while the lumped temperature state is a scalar, and discretisation failed with a shape mismatch.

This averages the surface temperature over the current collector(s) for the cooling term. It is a no-op for 0D (z_average of a scalar returns it unchanged). The 1+1D lumped model now builds and solves, and its temperature trajectory matches the 0D lumped result as expected.

[rtimms]

Looks good, thanks. Please drop the two-line comment above T_surf_av; the averaging is clear from the code.

[rtimms]

Thanks @chenyi6758-source, the fix is right. In the Timms2021 lumped formulation the cell has a single temperature, and the heat source and heat capacity are already averaged over the current collector, so the only term varying in z was T_surf (= T_amb(y, z, t)). Since h_total and A_cooling are uniform, averaging it gives exactly the same total cooling. I checked that 0D lumped is bit-identical to main, that 2+1D now builds too, and that 1+1D/2+1D lumped agree with x-lumped to within 0.5% of the temperature rise under uniform cooling.

Before this goes into 26.10.0.1:

1. Add a regression test. test_well_posed_lumped_thermal_1plus1D only checks the model symbolically, which is why this wasn't caught. Adding 1+1D and 2+1D "lumped" SPMe models to test_consistent_cooling in tests/integration/test_models/test_full_battery_models/test_lithium_ion/test_thermal_models.py, and asserting they match the x-lumped ones (e.g. assert 1e-5 > err(solutions["SPMe 1+1D"], solutions["SPMe 1+1D lumped"]), same for 2+1D), fails on main and passes here.
2. Drop the two-line comment above T_surf_av (as mentioned earlier); the averaging is clear from the code.
3. Optional, same root cause: "surface temperature": "lumped" with 1+1D/2+1D current collectors still fails, because thermal/surface/lumped.py uses the unaveraged "Ambient temperature [K]". Switching it to "Volume-averaged ambient temperature [K]" fixes both. Fine as a follow-up if you'd rather keep this PR minimal.