This was a cool question because it promoted thought about the DoM.
I raised a question How do HTML Parsers process untagged text which was commented generously by @sideshowbarker, which made me think, and improved my knowledge of the DoM, especially about text nodes.
Below is a DoM based way of finding candidate text nodes and padding them with 'p' tags. There are lots of text nodes that we should leave alone, like the spaces, carriage returns and line feeds we use for formatting [which an "uglifier" may strip out].