Part of #184 (Fountain import). Spec findings posted in the umbrella's research comment — read that first; this issue assumes those decisions.

What

Implement a Fountain parser in Rust that consumes a .fountain file string and produces a complete ScreenplayDocument (Film shape): meta, settings, content (ProseMirror JSON), and scene_cards.

Parsing pipeline (locked, mirrors reference implementation)

1. Boneyard pre-pass — strip every /* ... */ block (may cross double-line-breaks; this is the only Fountain syntax that does). Drop content per Fountain import (round-trip-safe) #184 scope. Increment a counter for the import summary toast.
2. Note extraction — find every [[ ... ]] (may span lines, but a fully blank line terminates). Record (text, anchor-position) and replace with a placeholder so line-based tokenisation doesn't see them.
3. Title page — top of file only. Recognised keys (case-insensitive, whitespace before : allowed): Title, Credit, Author, Authors, Source, Notes, Draft date, Date, Contact, Copyright. Multi-line values: continuation lines indented 3+ spaces or a tab. Title page ends at the first blank line followed by a non-key: line.
   • Map Title → meta.title, Author/Authors → meta.author, Credit → meta.author if blank else append, Draft date/Date → meta.draft_date, Contact → meta.contact, Copyright → meta.registration_number if empty else meta.extra, Notes → meta.footnote if empty else meta.extra.
   • Everything else (including unknown writer-coined keys) → meta.extra verbatim with original key spelling.
4. Body — paragraph-by-paragraph with the precedence below. Forced prefixes win over auto-detection.

Element precedence (per paragraph)

1. Forced prefix wins: ., !, @, >, ~, =, #. Note: .. is not a forced scene heading (must be a single . not followed by another .).
2. Page break: line is = × 3 or more and nothing else. (== is action; === 5 === is action.)
3. Centered: line starts > and ends <. (Beats forced-transition >.)
4. Scene heading auto: line begins with INT, EXT, EST, INT./EXT, INT/EXT, or I/E (case-insensitive) followed by . or space, with blank line before and after.
5. Transition auto: all-caps line ending in TO:, with blank line before and after. Also: FADE IN: at top of file (reference-impl convention; Fountain.js disagrees — we follow the reference).
6. Character auto: all-uppercase Latin line, ≥1 alphabetic char, non-empty next line, blank previous line. Lowercase inside parens is allowed (MOM (V.O.), HANS (on the radio)).
7. Parenthetical: context-dependent — line (...) immediately after Character or Dialogue.
8. Dialogue: context-dependent — line(s) after Character or Parenthetical, terminated by a fully blank line. A line with only trailing whitespace (e.g. two spaces) keeps the dialogue block alive (Fountain's two-space convention).
9. Action: fallback. Tabs and leading whitespace are preserved.

Mapping to Scriptty model

Malayalam handling (critical — see umbrella research comment)

• A Malayalam line cannot satisfy the all-caps character rule (caseless script).
• Auto-detect will treat unforced Malayalam character cues as Action — this is unrecoverable on import.
• The @ prefix is the only trustworthy signal: @രമേശ് → character node with text രമേശ്. Trust it regardless of caps.
• Document this in the import-summary toast: if zero @-prefixed cues are found in a file with substantial Malayalam content, surface a warning ("Fountain file appears to contain Malayalam but no @-prefixed character cues — characters may have imported as action. The sender should re-export with forced character prefixes.").

Implementation sketch

• New module src-tauri/src/screenplay/fountain_import.rs. Symmetric to the existing fountain.rs (export).
• Public surface: parse_fountain(input: &str) -> Result<(ScreenplayDocument, ImportSummary), String> where ImportSummary carries boneyards_dropped, sections_count, synopses_count, notes_count, dual_dialogue_count, scene_numbers_dropped, emphasis_stripped, plus warnings list.
• Hand-rolled parser — Fountain is simple enough that a nom/pest dependency would cost more than it saves. Plain string scanning + line iteration.
• Pipeline stages as separate functions: strip_boneyard, extract_notes, parse_title_page, tokenize_body, fold_tokens.

Test plan

• Round-trip fixtures: take 3–4 .screenplay files, export to Fountain via existing fountain.rs (after Fountain export: round-trip synopses, sections, inline notes, meta.extra #188 lands), re-import, diff. Must reach a fixed point after one full round.
• Hand-authored Fountain samples in src-tauri/tests/fixtures/fountain/:
  • title page only (standard + non-standard keys)
  • all element types
  • all forced types (., !, @, >, ~, =, #)
  • nested parentheticals, multi-paragraph action
  • Malayalam content with @-forced characters
  • Malayalam content without @ (must warn in summary)
  • sections at various depths (1–6)
  • synopsis on first scene + before any scene
  • boneyard (single-line + multi-line + mid-line)
  • dual dialogue
  • two-space-empty-line in dialogue (kept-alive blocks)
  • emphasis markers (stripped)
  • escape sequences (\\*, \\_)
  • page breaks (===, ====)
  • centered text
  • FADE IN: at top
  • all-caps action like BANG! followed by blank line (must stay action)
  • all-caps action ending TO: (auto-detected as transition — verify behavior matches spec, even if "wrong" by author intent)
• Unit tests on the lexer for ambiguous cases per the precedence list.
• Cross-check tricky cases against nyousefi/Fountain and wildwinter/screenplay-tools outputs (don't copy code).

Out of scope

• Inline emphasis → ProseMirror marks (data-loss tradeoff for v1)
• Lyrics as a dedicated node
• Custom scene numbers (#1A#) round-trip
• Headers/footers Highland-style directives ({{HEADER: ...}})