Replace regex-based markdown parsing with AST-based parsing

[jakehildreth]

Problem

Deck's inline/block parsing is a pipeline of regexes over flat strings:

• ConvertTo-SpectreMarkup applies bullet/bold/italic/color passes by regex with no structural context
• ConvertFrom-DeckMarkdown splits slides by --- matching with code-fence placeholders
• ConvertTo-CodeBlockSegments splits code blocks by non-greedy fence regex
• New-TableRenderable splits tables by | string ops

The converter receives bare strings with no context. It cannot distinguish a list item - 5% from a table cell containing - 5%, a fence-closing `````` from one inside a code sample, or a heading from heading-like text in code.

Concrete bugs caused by this architecture

• Formatted table content fails #25 — table cells require a sentinel guard so the bullet pass does not mangle leading - /*
• Markdown highlighting with nested code fence fails #9 — nested code fences terminate early (non-greedy regex, no fence-length tracking)
• Slide-type detection in Show-Deck.ps1 re-implements heading/content detection per branch (Allow H1/H2 content to mix with other content types #4, Enable Progressive Bullets in All Slides Types #5 root cause)
• No backslash-escape support anywhere (\*, \-) — CommonMark authors expect it
• Every new inline feature must slot between placeholder-replace and placeholder-restore and audit every other pass for conflicts

Proposed direction

Adopt a real markdown parser producing an AST, then render by walking typed nodes:

• Heading node → figlet renderer (level selects font) — fixes Allow H1/H2 content to mix with other content types #4 structurally
• List node → bullet renderer with progressive-reveal — fixes Enable Progressive Bullets in All Slides Types #5 structurally
• Table node → cell inline content rendered in cell context — no sentinel needed
• Code block node → TextMate with fence length from the parser — fixes Markdown highlighting with nested code fence fails #9's parsing half

Candidate engines: Markdig (.NET, CommonMark + GFM, loadable from PowerShell), or a PowerShell-native wrapper. Evaluate Markdig first — mature, MIT, single DLL, and Spectre.Console already proves the .NET-dependency pattern works for this module.

Non-goals for this issue

• Does not block Formatted table content fails #25 (fixing with sentinel guard under current architecture)
• Does not block Markdown highlighting with nested code fence fails #9 (partial fix possible with fence-length tracking regex)
• Not a rewrite commitment — spike first: prove Markdig loads and round-trips a representative deck

Relation to other issues

Supersedes the root cause of #4, #5, #9, #25. Those remain fixable independently in the interim.