Improvement: Cannot load datasets containing large numbers of 2D files

There are two related limitations in the current file loading workflow:

1. Cannot select a folder in “Upload from Local”

The file dialog does not allow selecting an entire directory.
While it is possible to select multiple .tif files manually (e.g., using Shift + click), selecting a folder directly is not supported.

2. Downstream features only allow single-file selection

For:

• Visualization
• Model training
• Worm error handling
• Other processing steps
  Only a single file can be selected at a time.

Impact

When working with datasets that contain thousands of 2D image files, these limitations make it effectively impossible to load and process the dataset efficiently.

Manually selecting thousands of files is not scalable, and loading them one-by-one is not feasible.

Expected behavior

• Allow selecting a directory containing image files.
• Allow loading a folder as a dataset.
• Support batch loading of multiple files across relevant features.
• Allow folder and subfolders to exist under our file management system, so that we can select a whole folder for training image/ask (right now images and masks are paired together in the same folder, so we see one image, one mask, ... This very hard for selecting files (if we get thousands of them) for training/error handling).

This would enable practical workflows for large 2D datasets.

No files, tests, or entry points are identified. Begin by tracing the “Upload from Local” selection flow and the selectors used by Visualization, Model training, and Worm error handling; define folder, subfolder, and batch-selection behavior before implementation. Done means large 2D datasets can be selected and processed without one-by-one file selection.