Auto-detect dataset formats and expand samples

[suhanpark]

Summary

Detect uploaded and sample dataset layouts, then configure the matching loader automatically.

Changes

• add structural detection for MobileBrick, Tanks and Temples, Hilti, YFCC IMB, Argoverse, Olsson, AstroVision, COLMAP, and generic image collections
• resolve detected paths relative to portable uploaded roots
• expand the curated GitHub sample catalog from 3 to 9 scenes and correct the Crane Mast image count
• add an Auto-detect (recommended) format option with a detection explanation in the UI
• cover the detector, uploads, sample catalog, and runtime integration with tests

Why

GTSFM sample datasets use different directory layouts. Requiring users to know the exact loader and path arguments made the Studio workflow unnecessarily error-prone.

Validation

• 47 focused visualization tests passed
• staged React/TypeScript frontend build passed
• git diff --check passed

Stack

Depends on #1127 and targets studio/runtime-reliability so this remains a focused, reviewable change.