[PaulHax]

Define raw and resolved classification boundaries

A track stores raw confidence pairs. The viewer can resolve a different display type from those
pairs when a hierarchy and filters are active.

This PR defines which features use the resolved display type. It also defines which features must
use the raw stored pairs.

Resolved display type

• Track and detection attribute filters use the same resolved pair as the viewer.
• Resolution uses hierarchy depth, checked types, and confidence thresholds.
• Suppression regions compare their type with the resolved display type.
• Lists, counts, and canvas code use the same suppression resolver.
• The suppression cache includes hierarchy and filter state.
• The cache also includes Prevent Cascade Types state.
• Attribute-name suppression keeps its existing attribute behavior.

Raw stored pairs

• Checked Types Only export compares checked names with raw pair names.
• Filtered DIVE JSON removes unchecked pairs from exported vectors.
• Full-archive export uses the same raw-name rule.
• Export filtering does not change stored tracks.
• Server Library labels use the raw pair with the highest confidence.
• A tie keeps the first stored pair.

The server cannot use viewer resolution for Library labels. It does not have each viewer's checked
types or confidence thresholds.

Related corrections

• Multicamera save keeps global attribute and filter definitions when the dataset uses named camera
  maps.
• The bulk type-update action appears only during multi-track editing.
• The bulk action cannot run with a hidden default during normal track selection.

Manual tests

Test data: download and extract
classification.zip. Paths
below are relative to dive-classification-test-data/.

Resolved attribute filters

1. Create the image-sequence dataset with the files in display/resolved-attribute-filters/, add
   the documented track and detection filters for juvenile-red-snapper, then change each filter
   type to fish.

Verify: Each track and detection filter uses the resolved type without a reload.

Resolved suppression regions

1. Create the image-sequence dataset with the files in display/resolved-suppression-regions/ and
   set the suppression region type to fish at the default 99% overlap.
2. Uncheck juvenile-red-snapper and red-snapper, then recheck both.

Verify: Track 2 disappears and returns without a reload.

Raw confidence-pair export

1. Create the image-sequence dataset with the files in export/raw-confidence-pair-filtering/;
   leave juvenile-red-snapper and fish checked, and uncheck red-snapper.
2. Export with Checked Types Only and inspect track 1 in the exported DIVE JSON.

Verify: The export contains only the expected raw pair names, and the source dataset does not
change.

Stack

This is PR 9 of 9. Previous: add DIVE KWCOCO classification support.

1. Allow devDependency imports in TypeScript spec files
2. Preserve warnings from every Desktop import file
3. Copy source metadata when creating a single-camera soft clone
4. Add hierarchical track classification
5. Centralize hierarchical classification changes
6. Replace mutable merged tracks with read-only projections
7. Make track lifecycle operations classification-safe
8. Add lossless DIVE KWCOCO classification support
9. Current — Define raw and resolved classification boundaries

This PR is stacked on hierarchical-classification-pr5-kwcoco.

Overall: 32 files, +723/-86. Commits: 90917889 through 25e427be.

[BryonLewis]

I added one fix that was basically making sure that the display and filtered exports respect the same 0 values. Before they didn't so you would have a mis-alignment between exporting and what you saw.