docs(examples): add uncertainty calibration plot example

[dimasd-angga]

Summary

Adds a small example script that produces an uncertainty calibration plot for the synthetic industrial world model, addressing the request in #17 for a beginner-friendly diagnostic. The script trains the demo ManufacturingRSSM with reduced defaults, runs the held-out test set through it, and saves a scatter of predicted quality uncertainty vs. absolute prediction error.

Changes

• Add benchmarks/toy-process-window/industrial-world-model/examples/plot_uncertainty_calibration.py: generates a dataset via generate_world_model_dataset, splits it with split_dataset, builds ManufacturingRSSM, trains for 10 epochs with train_world_model, then iterates the test ManufacturingSequenceDataset to collect state_mean, state_std, and target next_state on the QUALITY_INDEX dimension.
• The same script computes absolute error, prints a short numeric summary (sample count, mean absolute error, mean predicted uncertainty, and the uncertainty/error Pearson correlation), and writes uncertainty_calibration.png next to itself using matplotlib.

Testing

Manual run only: cd benchmarks/toy-process-window/industrial-world-model && python examples/plot_uncertainty_calibration.py. The script trains the demo model, prints the summary, and writes uncertainty_calibration.png in the examples/ directory. No automated tests were added — the issue asks for an example/plot, not a code path under test.

Notes

• Defaults (40 episodes, 30 steps, 10 epochs, hidden 128 / latent 32) are intentionally smaller than run_demo.py to keep first-run time low; bump them for a smoother plot.
• Matplotlib is imported lazily inside main() since it is only needed at plotting time; it is already a transitive dependency via imwm's visualization.py.
• No source modules under imwm/ were modified.

Closes #17

[programmablemanufacturing]

Thanks for the contribution — this is exactly the kind of beginner-friendly diagnostic example I was hoping for.

Before I merge, could you please reformat the Python file with standard line breaks and indentation? The raw file appears to have many statements collapsed onto the same lines, so it may not run correctly as Python.

Could you also confirm that this command works from the industrial-world-model directory?

python examples/plot_uncertainty_calibration.py

Once the formatting is fixed and the script runs, I’d be happy to review again.

[dimasd-angga]

Pushed an update addressing the review feedback.

Commit: docs(industrial-world-model): document uncertainty calibration example