fix(jax): defensive pytree dedup in imaging/interferometer analyses

[Jammy2211]

Summary

Same defensive pattern as the PyAutoGalaxy companion PR — additionally cross-populates `Tracer` from autofit's `_REGISTERED_INSTANCE_CLASSES` since lens models always carry one, making `Tracer` a regular source of duplicate-registration whenever both the autofit walker and the analysis registration path run.

Repro

Same root cause as autogalaxy companion (multi/start_here.py in autogalaxy_workspace during the 2026.5.29.2 release):

```
ValueError: Duplicate custom PyTreeDef type registration for
<class 'autoarray.dataset.dataset_model.DatasetModel'>.
```

The autolens analyses register the same class set plus `Tracer`. Once `autofit.jax.pytrees.register_model` walks a lens model, both `DatasetModel` and `Tracer` get registered via autofit's path — neither set lands in autoarray's `_pytree_registered_classes`, so the next `fit_from` call's registration step asks JAX to register them again.

Fix

For both `imaging/model/analysis.py` and `interferometer/model/analysis.py`:

1. Module-level `FIT*_PYTREES_REGISTERED = False` flag.
2. Cross-populate `(DatasetModel, Tracer)` from autofit's set into autoarray's.
3. Set the flag at the end.

Paired with

• PyAutoGalaxy companion PR (same fix)
• PyAutoBuild PR (release.yml `run_scripts` env block)

Test plan

• `pytest test_autolens/imaging/model test_autolens/interferometer/model` — passed (unit tests don't exercise JAX path)
• CI on this PR
• Next `autobuild pre_build` dispatch

🤖 Generated with Claude Code