Slate v2 cross-block delete should reuse the merge seam

What happened

Once Editor.before(...), Editor.after(...), and Transforms.move(...) could cross supported top-level block boundaries, delete(...) still lagged on that seam.

The tempting lie was to special-case cross-block deletion with a separate tree rewrite path.

That would have duplicated block-join semantics that mergeNodes(...) already owned.

What fixed it

The honest follow-on kept the delete helper small:

1. delete edge text on the start/end sides when needed
2. reuse mergeNodes(...) for the adjacent block join
3. collapse the selection back to the computed start boundary

That keeps adjacent cross-block delete aligned with the same merge contract the core already proves elsewhere, whether the range started collapsed or was already explicit.

Reusable rule

For Slate v2 adjacent cross-block deletion:

• if the behavior is an adjacent block join, reuse the merge seam
• do not fork a second block-join implementation into delete(...)

If block-boundary delete and block merge have separate semantics, one of them will drift.

Sweep guard

When reviewing core editor methods for this regression class, hunt for:

• private structural helpers that manually merge, split, lift, wrap, or unwrap where a primitive seam already owns the behavior
• optional transaction parameters that fall back to editor.apply
• preflight checks that read live selection before transaction target resolution
• direct Transforms.* use in the core mutation surface

The correct fix is usually to route the method through the owned primitive seam, not to make the local helper smarter.