double-encoding ampersands or angle brackets when repopulating editor

[RwwL]

I want to store our editor content in a custom format to, among other reasons, allow rendering the content in other places without a Tiptap instance and having flexibility in the level being applied to heading elements. For top-level nodes other than headings, we're fine with storing the plain HTML. So, here's an example of what we're storing for a given document:

const savedEditorData = [
  {
    type: "heading",
    value: "Testing 1, 2, 3",
    options: {
      attrs: {
        level: 1
      }
    }
  },
  {
    type: "text",
    value:
      '<p>Paragraph after H1.</p>'
  },
];

My approach to repopulating the editor with saved content has been to initialize the editor with content: '', then loop over the server content, calling insertContent for each top-level object. If the object is of type: "heading", then I use the generateHTML util from @tiptap/core to generate the full HTML for the heading node, and if it's not then I just use insertContent(obj.value).

So when we restore saved content to repopulate the editor, we loop over that saved-data array and here's what we do if the object type is heading:

  editor.chain().focus('end').insertContent(
    generateHTML({
      type: 'doc',
      content: [
        {
          type: 'heading',
          attrs: {
            level: obj.options.attrs.level,
            textAlign: obj.options.attrs.textAlign,
          },
          content: [
            {
              type: 'text',
              text: obj.value,
            },
          ],
        },
      ],
    }, editorExtensions),

The only issue I'm having with this is when content in a header (but not paragraph/list/other nodes) contains &, <, or >: because I called the generateHTML utility to store the header's content, and then I'm calling generateHTML on it again, I'm winding up with double-escaped entity bugs that render as & in the editor.

I've tried not calling generateHTML here, just passing that object directly to insertContent, but that doesn't solve it.

Anyone have thoughts on a better (or Tiptap-"correct") way to manage this than simply doing a replace on the stored HTML right before I restore it into the editor? Since the editor is clearly escaping incoming content, I don't think I'm exposing the editor to XSS attacks by doing that. I checked to see if type: 'html' is supported instead of text in the nested header content in the insertContent arguments above, but it's not.

I've looked the parseOptions option that can be passed to insertContent but don't see anything that I think might help here.