Translation issues (duplication and missing words) for german source with html encoded <

First of all: thanks for your work!

I stumbled into the following issues:

using the [example](https://github.com/argosopentech/translate-html) with `from_code = "de"` and `to_code = "en"`:

`html_doc = """

<

"""`

translates to:
```HTML

<= <= < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < <

```

expected translation: `

<

`

and

`html_doc = """

< Fehler

"""`

translates to:
`<

<=

`

expected translation: `

< error

`

Start with the linked example and reproduce both German-to-English inputs containing the HTML-encoded less-than entity. Trace the translation path that handles these inputs and compare its output with the expected HTML shown in the issue; done means both examples preserve the entity and translate "Fehler" to "error" without duplication.

Written by the indexing model from the issue text.