feat: add nested field columns

[MeisQuietude]

What changed

• Make the grid projection an ordered list of schema paths selected through a searchable tri-state schema tree; keep right-click Flatten and Unflatten plus Peek Promote to column as shortcuts over that projection.
• Preserve nested-column filtering, sorting, statistics, pinning, copy, export, duplicate-name handling, empty-name handling, and sparse Parquet leaf reads for arbitrary projected struct leaves.
• Infer a bounded, sample-derived tree for logical JSON columns and support filters and sorts on sampled or manually entered object keys and array indices through cancellable scans.
• Keep Peek stable while selecting or copying fields, size it for the active value, expose field actions through context menus and keyboard shortcuts, and make manual resize visible and predictable.
• Keep Parquet field paths and JSON paths as separate wire contracts, with regression coverage for deep and wide schemas, nulls, unusual names, numeric boundaries, cancellation, and source transitions.
• Closes Nested fields as first-class columns: flatten, path addressing, statistics #83.

Why

Users need to select the nested leaves they actually work with, while JSON fields should remain queryable without materializing them as grid columns.