keyed 0.2.0

New Features

• watch() / unwatch(): Mark keyed data frames as "watched" so dplyr verbs auto-stamp before executing. Turns drift detection from a manual ceremony into an automatic safety net.
• check_drift() now returns cell-level reports via keyed_diff when keys match. Falls back to structural comparison when keys differ or are lost.
• diff.keyed_df(): Cell-level comparison using key columns to align rows. Reports added, removed, and modified rows with per-column change detail.
• stamp(.silent): Suppress cli output during auto-stamping.
• list_snapshots() gains a size_mb column.
• compare_structure(), compare_keys(): Structural comparison helpers.

Breaking Changes

• commit_keyed() soft-deprecated in favour of stamp().

Internal

• Snapshot cache stores full data frames (not just hashes), enabling cell-level drift comparison.
• Cache reduced from 100 to 20 entries with 100 MB soft memory cap.
• All dplyr methods propagate snapshot references and watched state through transformations.

Full changelog: v0.1.1...v0.2.0