feat(parser): opt-in typographic smart punctuation

[DemchaAV]

Stacked on #41 → #40 → #39 — merge in order; this is the top of the stack.

Why

Tier 3 typographic polish from the roadmap: print-quality documents want real dashes, ellipses and curly quotes. Opt-in because GitHub does not smart-quote — default output stays byte-identical.

What changed

• MarkdownComposer.builder().smartPunctuation(true) enables flexmark-ext-typographic on the parser.
• Mapper handles TypographicSmarts (— – …) and TypographicQuotes (curly marks around mapped children, so "a **bold** word" keeps its bold) by riding the existing HTML-entity decoder; curly-quote/guillemet entities added.
• Code spans and code blocks stay verbatim (flexmark guarantee, pinned by test).
• Note: superscript/subscript from the same roadmap line was not built — verified engine-gated (no baseline shift on engine text runs); recorded in the private roadmap.

Verification

./mvnw clean verify javadoc:javadoc → BUILD SUCCESS, 173 tests, 0 failures. New: replacements on; off-by-default literal; code immunity with the flag on; formatting-inside-quotes.

[DemchaAV]

Blind cold review: approve with minor changes — all landed (amended into b985d6b): the false byte-identical CHANGELOG claim is fixed (the entity-decoder default-mode change is now its own honest Changed bullet); apostrophes / «guillemets» / spaced-ellipsis conversions documented in the builder Javadoc + CHANGELOG + README; null-guard on getTypographic* (NPE was reachable via render(Document) with hand-built nodes); +3 tests (apostrophe path, inside-::: replacements, model-level bold-inside-quotes, entity-leak guard). Reviewer had byte-verified all six new entity codepoints and proven the entity table complete for every string the extension can emit. Suite: 178 green.