Incorrect HtmlBlockMarkerBlock token detection

1. Input `` or ` text ` is parsing like a single `HTML_BLOCK_CONTENT`.
2. Input `A ` is parsing as TEXT(`A`), WHITE_SPACE, HTML_TAG(``).
3. Input `A text ` is parsing as TEXT(`A`), WHITE_SPACE, HTML_TAG(``), WHITE_SPACE, TEXT(`text`), WHITE_SPACE, HTML_TAG(``).

The first inputs should work the same way as samples 2 and 3.

Start by locating the HtmlBlockMarkerBlock token-detection code and reproduce the four inputs from the issue. The work is done when the first two inputs tokenize consistently with samples 2 and 3, rather than as a single HTML_BLOCK_CONTENT token.

Written by the indexing model from the issue text.