Remove range row points at expanded multiblock delete not another Android quirk

What happened

The Android remove row was promoted into a direct proof surface.

Current setup:

• prepare a model-owned expanded selection from block 0 into block 2
• try to delete the range
• then backspace the rest

Useful findings:

• the prepared selection is real in the model
• the DOM selection is also real on Chromium
• printable replacement over that selection works
• Appium backspace deletes the selected content, but leaves three empty blocks
• Playwright desktop delete/backspace is inert on that prepared range

That means this is not just “another Android weirdness” row.

Hard read

The remaining problem is clustering around one deeper lane:

• expanded multi-block delete semantics
• delete ownership across browser key paths
• convergence/merge after the deletion

This is a core structural editing problem, not a one-off mobile proof hole.

Reusable rule

When a remaining Android/manual row shows:

• real prepared selection
• inconsistent delete behavior across transports
• leftover empty blocks instead of converged structure

do not keep attacking it as isolated platform parity.

Promote it into a structural deletion tranche:

1. expanded multi-block deleteFragment
2. browser delete key ownership (Backspace / Delete)
3. post-delete block convergence