Add LaTeX, super/subscript, pull quotes, callouts, and Markdown docs

[MattFisher]

Builds on the from_markdown work from #61, adding several Markdown constructs that map to Substack nodes, plus a documentation page. Node shapes were confirmed by round-tripping editor-authored drafts through the live API.

Closes #63

New Markdown features

• LaTeX math — inline $…$ (→ latex node) and block $$…$$ (→ latex_block), via mdit_py_plugins.dollarmath.
• Superscript (^x^) and subscript (~x~) inline marks. Subscript's single-tilde syntax coexists with ~~strikethrough~~.
• Pull quotes and callouts via fenced-container syntax (:::pullquote / :::callout), since they have no native Markdown equivalent.

Tables: intentionally not supported

An earlier iteration added a table node, but Substack has no table renderer or editor UI — the node stored but never displayed. It's been removed to avoid producing silently-broken drafts. docs/markdown.md documents this and points users at Datawrapper embeds for tabular data.

Docs

New docs/markdown.md documents every supported construct with examples, plus what's intentionally unsupported. Linked from the README's Markdown section.

Dependencies

The sub/superscript plugins require mdit-py-plugins >= 0.5; the constraint (>=0.5,<0.7) and lockfile are updated accordingly (resolves 0.6.1).

Testing

• Unit tests cover each new construct in test_from_markdown_features.py.
• The end-to-end fixture and golden file were regenerated from a live round-trip, so every feature here is verified against the real Substack API (RUN_SUBSTACK_E2E=1).

[ma2za]

Thanks for adding these Markdown features. I reproduced two edge cases against the PR head:

1. Ordinary paired dollar signs are parsed as math. Revenue grew from $5 million to $10 million this year. becomes text, latex, text.
2. Labeled display math is dropped. For $$ ... $$ (mass-energy), the plugin emits math_block_label, but the renderer only handles math_block.

These regression tests currently both fail:

def test_currency_dollar_signs_are_not_parsed_as_latex(self):
    post = make_post()
    markdown = "Revenue grew from $5 million to $10 million this year."
    post.from_markdown(markdown)

    paragraph = body(post)[0]
    assert [node["type"] for node in paragraph["content"]] == ["text"]
    assert paragraph["content"][0]["text"] == markdown


def test_labeled_latex_block_is_preserved(self):
    post = make_post()
    post.from_markdown(
        "$$\nE=mc^2\n$$ (mass-energy)\n\nAfter the equation."
    )

    blocks = body(post)
    assert [block["type"] for block in blocks] == ["latex_block", "paragraph"]
    assert blocks[0]["attrs"]["persistentExpression"] == "E=mc^2"

The focused run reported 2 failed, 26 deselected.

[MattFisher]

Good catches! Both fixed. The dollarmath plugin is now configured with allow_space=False, allow_digits=False (Pandoc-style delimiters), so paired currency amounts stay plain text, and the renderer now handles math_block_label the same as math_block, discarding the label since Substack has no equation labels. Both of your regression tests are added to the suite and pass.