Estimate Scenario sizes without materializing unloaded datasets

Start by tracing Scenario's read_only_dataset_resolution() and how DatasetConfiguration metadata, CentralMemory, and provider population metadata are used for sizing. Verify that default and configured estimates remain read-only, use caps and available metadata without materializing missing datasets, and preserve the existing estimate schema.

Is your feature request related to a problem? Please describe.

On a fresh backend, Scenario default estimates use read_only_dataset_resolution() to prevent estimation from fetching and persisting datasets. If a named default dataset is not already present in CentralMemory, resolution raises DatasetConstraintError, so the catalog reports unavailable population counts and estimates until another operation loads the dataset.

This was identified while reviewing #2730. The timeout and concurrency hardening in that pull request should remain side-effect free and does not address dataset sizing semantics.

Describe the solution you'd like

Allow default and configured Scenario sizing to use DatasetConfiguration metadata, including max_dataset_size and configured caps, together with provider population metadata or count APIs. Capped defaults should usually produce a planned size without fetching and materializing every seed.

The estimate path should remain read-only: it must not fetch and persist missing datasets as a side effect.

Describe alternatives you've considered, if relevant

• Auto-fetching missing datasets during estimation would make more estimates available, but violates read-only preview behavior and can introduce expensive persistence side effects.
• Continuing to require datasets to be preloaded preserves current behavior but leaves fresh-backend catalog estimates unavailable.

Additional context

This aligns with the approved standalone Scenario GUI follow-up plan's dataset-cap semantics and provenance work. The implementation should preserve the existing estimate schema until the structured run-size and dataset-provenance follow-ups intentionally evolve it.