Slate v2 before/after should reuse mixed-inline offset projection

What happened

The first Editor.before(...) / Editor.after(...) slice only walked inside one text node.

The next honest follow-on was walking across sibling text leaves inside one supported top-level text block, including inline descendants.

The tempting lie was to bolt more branchy cursor math straight into editor.ts.

That would have duplicated logic the core already had for mixed-inline text offset projection.

What fixed it

The honest follow-on moved point stepping into the core and reused the existing mixed-inline helpers:

1. project the current point to a block-text offset
2. add or subtract the distance
3. project back into a concrete point path/offset

That makes before(...) / after(...) agree with the same block-text model already used elsewhere in the core.

Reusable rule

For Slate v2 location walking:

• reuse the core's mixed-inline offset projection when the block model already exists
• do not duplicate tree-walking math in editor.ts
• broaden block-local walking before attempting cross-block walking

If before(...) and after(...) compute mixed-inline steps with their own handmade tree crawl, they will drift from the rest of the engine.