Auto completion `textEdit`.

We are currently returning atomic labels only because the clients are only taking the last part of a composite name into account.
@Kha pointed out that the LSP spec contains a discussion that is likely relevant

    /**
     * A string that should be inserted into a document when selecting
     * this completion. When `falsy` the label is used as the insert text
     * for this item.
     *
     * The `insertText` is subject to interpretation by the client side.
     * Some tools might not take the string literally. For example
     * VS Code when code complete is requested in this example
     * `con<cursor position>` and a completion item with an `insertText` of
     * `console` is provided it will only insert `sole`. Therefore it is
     * recommended to use `textEdit` instead since it avoids additional client
     * side interpretation.
     */
    insertText?: string;

We're not even using insertText?, but this suggests that we should "upgrade" straight to textEdit if we want to make depth > 1 work.

The issue names no repository files, tests, or entry points; start by locating where completion items are assembled and compare the current atomic-label behavior with the LSP specification's textEdit discussion. Done means completion for names with depth greater than one inserts the intended composite text rather than only its final part.