patchDocument: incorrect output when patches have no text

[SebKranz]

I encountered weird results when using the document replacer:

• the SPLIT_TOKEN (ɵ) appeared in the final document.
• Placeholders within the same paragraph appeared out of order.

The following seems to be happening:

1. I was instantiating a TextRun with an empty-string

   {
     type: PatchType.PARAGRAPH,
     children: new TextRun({ text: "" }),
   }

2. since I was calling the constructor with an object, rather than a plain string (new TextRun("")), the TextRun doesn't contain a Text child.
3. the replacer inserts the SPLIT_TOKEN, and then crashes when trying to access the now missing elements property here: https://github.com/dolanmiu/docx/blob/c6bb2556419f49f0ea42e774a20f5fa5f7450350/src/patcher/replacer.ts#L70C1-L74C1
4. because patchDocument catches all errors, the merge continues and future runs on the same paragraph use the incorrect split-token.

I think there are three problems to be addressed:

• the code for detecting when all patches are found is masking all unintended errors. This should be communicated in the returned value, not an error.
• the replacer should handle empty Runs.
• there should be an option for controlling the white-space TextRun({ space: "preserve", text: "..." }) EDIT: I mistakenly thought the w:t was missing due to the formatter

This issue does not occur when:

• keepOriginalStyles is off
• we instantiate the Run with a string argument (new TextRun("...")), since that will run through a different code path in the constructor.