Fix list/blockquote spacing cascades; add nested indent + bullet hierarchy

[thesmallstar]

Summary

• Deeply nested lists (3+ levels) and blockquotes were producing 2+ blank lines between blocks because each *_close handler emitted \n independently with no awareness of the cascade. Nested bullets also had no indent.
• This PR fixes both: adds proper indent + bullet hierarchy for nested lists, and a \x02 STX sentinel + cap regex in render() to collapse multi-newline runs from cascading closes down to a single blank line — without touching content inside fenced code blocks.

What changed

src/slackify_markdown/slackify.py

• _list_depth state drives a 4-space indent per nest level in list_item_open.
• _BULLETS_BY_DEPTH = ("•", "◦", "▪") — matches Slack's native nested-list glyphs.
• Close-handlers (heading_close, paragraph_close, bullet_list_close, ordered_list_close, blockquote_close) emit NEW_LINE = "\x02" instead of literal \n.
• render() caps runs of 3+ sentinels to 2 with one regex, then replaces sentinels with real \n. Code-block / fence handlers emit real \n directly so their content is preserved verbatim (including multi-blank-line content).
• slackify() scrubs \x02 from user input so a literal STX in a Markdown source can't collide with our cap machinery.

tests/test_convert.py

• 10 new edge-case tests: deeply nested bullets (4+ levels), mixed ordered/unordered nesting, code blocks with special chars and blank-line preservation, loose lists, blockquote+inner-list, all 6 heading levels, multi-blank-line collapse with code-block preservation, links with nested formatting + URL ampersands, inline-code + Slack mentions, and the STX-sentinel input scrub.
• 1 new full-document integration test (test_full_document_with_all_patterns) that exercises ~everything together.
• Updated test_complex_markdown expectations for the cleaner spacing.
• 60 tests pass.

docs/architecture.md (new)

• Pipeline overview, format mappings, full rationale for the sentinel/cap mechanism.
• Trail of alternatives considered for the sentinel char (regex on real \n, split on ```, PUA , Unicode noncharacters, NULL \x00, STX \x02) and why STX won — same convention as python-markdown.
• Known limitations (multi-paragraph items in lists, code-blocks-in-lists, hardbreak continuations, multi-line blockquote prefix) with note that AST-walker migration would resolve all four. Linked to Migrate renderer to AST tree-walker (resolves cascade + indent limitations) #19.

Why STX over alternatives

Choice	Verdict
\n{3,} regex on real newlines	rejected — eats blank lines inside code blocks
 Private Use Area	rejected — Unicode FAQ explicitly warns this collides with real PUA usage
� Unicode noncharacter	rejected — same valid-codepoint risk
\x00 NULL	rejected — Python source files cannot contain literal NULL; shells / argv / os.exec* reject NULL
\x02 STX	chosen — battle-tested by python-markdown; ASCII-safe in source, shells, JSON, filesystems

Test plan

• PYTHONPATH=src python3 -m pytest tests/ -v — all 60 pass
• Manual Slack render via webhook of .venv/sample.txt (LLM-generated content with 2-level nested lists + links) — clean single-blank-line separation everywhere
• Verified user-typed \x02 in input does not corrupt output
• Verified \n\n\n\n inside a fenced code block is preserved
• Reviewer: visually confirm rendered Slack output looks clean for a typical assistant message

Follow-ups

• Migrate renderer to AST tree-walker (resolves cascade + indent limitations) #19 tracks migration to an AST tree-walker renderer (resolves the four known limitations above and obsoletes the sentinel + cap mechanism)