fix: broadcast the average SEI film overpotential with a size distribution

[shaurya20335]

Description

With "SEI film resistance": "average" and "particle size": "distribution", SPMe, DFN and Newman-Tobias fail to build:

import pybamm
pybamm.lithium_ion.DFN(
    {"particle size": "distribution", "SEI": "constant", "SEI film resistance": "average"}
)

DomainError: children must have same or empty domains, not ['negative particle size'] and ['current collector']

In BaseKinetics, the average film drop eta_sei = -j_tot_av * L_sei_av * R_sei lives on the current collector. When the size distribution is resolved through the electrode, eta_r is on the particle-size domain with the electrode as its secondary domain, so eta_r += eta_sei cannot broadcast across the missing electrode level. MPM is not affected because its eta_r already has the current collector as its secondary domain, and "distributed" is not affected because its eta_sei is on the electrode.

eta_sei is now broadcast onto the electrode in that case. Nothing changes for models that already built: MPM and DFN without a distribution give bit-identical voltages with the average film.

With the fix (OKane2022, "SEI": "constant", 50 nm film, 5 A, voltage at 600 s):

model	no film	average film	distributed film
DFN with size distribution	3.80096 V	3.78611 V	3.78574 V
SPMe with size distribution	3.80342 V	3.78859 V	3.78773 V
DFN (reference, no distribution)	3.80921 V	3.79431 V	3.79304 V

The average and distributed films differ by about 1 mV with a size distribution, as they do without one (1.5 mV for the DFN).

I found this while checking #5841 across model options.

Type of change

Bug fix, with an entry under ## Bug fixes in CHANGELOG.md.

Important checks:

• No style issues: ruff 0.16.10 format and check, and comment-slop, pass on the changed files
• Tests added that prove fix is effective: test_well_posed_psd_sei_film_resistance (unit, in the shared lithium-ion base tests) builds each model with a size distribution and both film options. It fails on main for SPMe, DFN and Newman-Tobias and passes with the fix.
• All tests pass: the full unit suite (4,852 passed, 89 skipped) and the full integration suite (518 passed, 29 skipped) pass locally on macOS with Python 3.12. I am relying on CI for the notebooks, example scripts and other platforms.
• The documentation builds: not run locally, no docs changes
• Code is commented for hard-to-understand areas

[rtimms]

Looks correct, thanks. I checked it together with #5841 (SEI film heating) and #5875: every size-distribution model builds with no, average or distributed film, and with #5841 the "average" film's extra heat matches its extra electrical loss to 0.13% (SPMe) and 0.01% (DFN) with a size distribution.

Heads-up on #5818, which deprecates relying on options set from other options (the test suite turns that warning into an error): test_well_posed_psd_sei_film_resistance fails with it for all 4 models, because the film options set "total interfacial current density as a state": "true" by default. Whichever of this PR and #5818 merges second needs to pass that option explicitly.

[shaurya20335]

Thanks for checking it with #5841 and #5875, @rtimms. test_well_posed_psd_sei_film_resistance now passes "total interfacial current density as a state" explicitly ("false" for the average film and "true" for the distributed film), so it should not depend on the order this and #5818 merge in.