[Migration] Stop pymc's forward sampler rejecting the aDDM p_outlier TensorConstant

Part of #1305. Parent: #1306.

⚠️ **Not in [bambi#1002](https://github.com/bambinos/bambi/pull/1002)'s notes** — found by re-triage after #1310 unmasked the aDDM predictive paths.

aDDM posterior predictive sampling fails inside bambi's new in-sample `predict`:

```
TypeError: ('Constants not allowed in param list', TensorConstant(TensorType(float64, shape=()), data=array(0.05)))
bambi/backend/pymc/model.py:477: in _predict_in_sample
pymc/sampling/forward.py:387: in compile_forward_sampling_function
```

bambi 0.20 compiles the forward sampler over the model's inputs (`pm.sample_posterior_predictive` on a cloned model), and the aDDM graph carries `p_outlier` as a `TensorConstant` where pymc expects a shared variable or input. aDDM only — the SSM predictive path is fine.

### Tasks
- [ ] Find where the aDDM distribution bakes `p_outlier` in as a constant instead of a model input (compare with the SSM path, which passes)
- [ ] Route it through the same mechanism as the other fixed parameters, or wrap it so pymc treats it as an input
- [ ] Remove the R12 xfail marks

### Impact
3 test ids across 3 files: `tests/addm/test_addm_ppc.py`, `tests/addm/test_addm_continuation.py`, `tests/addm/test_addm_cartoon.py` (1 each).

Details: `design/bambi-dev-upgrade-test-failures.md` (R12), `design/bambi-migration-fix-plan.md` (F11).

Start by comparing the aDDM distribution's handling of p_outlier with the passing SSM path, using bambi/backend/pymc/model.py:477 and the migration notes in design/bambi-dev-upgrade-test-failures.md (R12) and design/bambi-migration-fix-plan.md (F11). Run the three affected tests: tests/addm/test_addm_ppc.py, tests/addm/test_addm_continuation.py, and tests/addm/test_addm_cartoon.py; done means the forward sampler no longer raises the TensorConstant error and the R12 xfails are removed.

Written by the indexing model from the issue text.