[nk-batman21]

With the fenced-code-blocks extra, a line inside a code block that ends with the opening backticks closes the block there, wherever the backticks sit on the line. This happens whenever the code itself contains a Markdown fence, for example a Python function whose docstring has an indented example:

```
def greet():
    """Say hello.

    ```
    greet()
    ```
    """
```

Some *text* between the blocks.

On master the block ends at the indented ``` line. greet() becomes a separate code block, the real closing fence then opens a new block that swallows "Some text between the blocks.", and code and prose swap places for the rest of the document. A code line such as Wrap code in ``` does the same.

This is a regression from #418, which changed the closing fence from ^```[ \t]*\n to \1[ \t]*\n and dropped the ^. This puts the anchor back and lets the closing fence be indented up to 3 more spaces than the opening fence or be longer than it. Both of those already closed a block by matching partway through the line, and they no longer leave a stray whitespace or blank line at the end of the code. A closing fence indented 4 or more spaces past the opening one no longer closes the block, as in CommonMark.

Ran: make testone on Python 3.9 to 3.14 (304 tests pass) and make testredos; the new fixture fails without the change.
Checked: the docstring example is fixed with Pygments, highlightjs-lang and safe mode too. Fences inside list items, blockquotes, footnotes and admonitions render as before, and the nested fences from #327 still work.

[nicholasserra]

Looks reasonable, merge master in or rebase and should be able to merge if passes CI thanks