Nested clipboard fragments must merge into active text block

Problem

Lexical regression 1384 is named for insert nodes, but the portable behavior is native clipboard insertion inside a code block. Copying text from a nested code line and pasting it inside another code line should insert text into the active line, not create a new code block sibling.

Symptoms

• The package proof inserted a separate code-block before the target code-block.
• The browser proof showed six top-level blocks instead of five.
• The visible pasted text landed in a tiny standalone code block before the original code block.

What Didn't Work

• Treating the row as a generic insert-nodes API test. The user-visible bug was clipboard fragment fitting.
• Fixing the code-highlighting example with a paste handler. That would hide a core insertion bug and duplicate editor policy in product code.
• Asserting only clipboard text. The broken path was structural, not text extraction.

Solution

Add a package contract for inserting this fragment shape:

[
  {
    type: 'code-block',
    children: [{ type: 'code-line', children: [{ text: 'Add' }] }],
  },
]

into a collapsed selection inside an existing code-line.

Then update insertFragment nested fitting so a single nested text-block child from the same structural container merges into the active nested text block. The operation replaces the container's children in one logical operation and places selection after the inserted text.

Finally, add a browser proof in the code-highlighting example that uses native desktop clipboard shortcuts and asserts:

• the top-level block count stays unchanged;
• pasted text lands in the current code line;
• the following code line stays intact;
• the code block language metadata stays attached.

Why This Works

Slate clipboard fragments preserve ancestor structure. A partial copy inside a code line can therefore arrive as code-block -> code-line -> text, even though the intended paste target is already inside a compatible code-line.

The nested fitting path already knows how to replace the shared structural container. The missing branch was the single-child case where the copied container has exactly one compatible text-block child. Merging that child's children into the active text block preserves the surrounding code block and keeps the paste local to the active line.

Prevention

• For harvested clipboard rows, assert structure after paste, not just plain text.
• Add a package-level fragment shape proof before browser-only fixes when the failure inserts the wrong Slate node structure.
• Keep example paste handlers out unless the behavior is product policy rather than generic fragment fitting.

Related Issues

• Lexical codeblock harvest rows need data transfer boundaries