Add Twig and SCSS extractors

[amslezak]

What

Adds extractors for Twig templates and SCSS/Sass stylesheets. Neither extension appeared in any FileType set, so both were skipped at detection — rg twig over the package returns nothing today. On a Drupal or Symfony project that means the entire template-composition layer is missing from the graph, and on any Sass project the stylesheet module graph is too.

Both extractors follow extractors/blade.py: pure regex, no new dependency. Routing them through FileType.CODE rather than the LLM document path means they cost no tokens and produce EXTRACTED-confidence edges with precise relation names.

Two commits: the extractors, then a resolution fix they exposed.

1. The extractors

graphify/extractors/twig.py resolves references to real files rather than leaving stubs, so templates link to templates:

• Drupal SDC — {% include 'mytheme:card' %}, resolved via the nearest ancestor <provider>.info.yml plus an index of components/**/<name>/<name>.twig.
• Plain paths and Symfony — {% extends 'layout.html.twig' %} and @Namespace/path.html.twig, resolved against the referring file's directory and any enclosing templates/ root.

Relations: includes, embeds, extends, imports_macro, uses_template, defines_block, attaches_library. Covers the {% tag %} form, the {{ include('x') }} function form, and {% from 'x' import y %}.

graphify/extractors/scss.py covers three graphs:

• Modules — @use / @forward / @import through Sass's partial (_name.scss) and index (name/_index.scss) conventions. Handles @import 'a', 'b' lists; leaves sass:math a stub.
• Mixins — @mixin / @include keyed on the bare name, so a namespaced use (@include mx.button-reset) links to its definition instead of orphaning.
• Design tokens — --token: declarations and var(--token) references.

Both follow the #2195 pattern: a resolved reference emits the edge only, stamped with target_file, and mints no stub node — a second node carrying the same id would be a ghost duplicate of the real file node.

Comment stripping

Both strip comments first, via a newline-preserving blank pass (_blank_spans in base.py) so reported line numbers stay accurate.

This is load-bearing. Component docblocks routinely carry a usage example:

{#
 * Usage:
 *   {% include 'mytheme:card' with { title: 'Hello' } %}
 #}

Read naively that gives the card component an includes edge to itself — a self-loop, which diagnose_extraction flags. Same for a commented-out // @use 'ghost';.

2. Keeping composition edges alive through id disambiguation

Validating against a real theme surfaced a resolution bug that isn't Twig-specific.

A Single Directory Component keeps card.twig beside card.scss. Both collapse to the same card file id, so _disambiguate_colliding_node_ids salts them apart by path. An {% include %} edge from a third template carries neither salt's source_key, so the (target, edge_source_key) lookup misses and the edge dangles on the now-retired card id.

That is the #1475 C/ObjC header failure exactly — foo.h beside foo.c, with the import arriving from a third file. This mirrors that fix: repoint template-composition edges to the .twig variant, since an include always targeted the template.

It also honors the target_file stamp for those relations. The gate was ("imports", "imports_from", "re_exports"), so a stamped includes edge fell back to the importer's own source_key.

Measured on a 141-template / 109-stylesheet Drupal theme:

	before	after
includes dangling	198 / 234	0 / 234
embeds dangling	14 / 14	0 / 14
extends dangling	1 / 1	0 / 1
Sass imports resolved	—	84 / 84
self-loop edges	—	0

The only unresolved Twig references left are base-theme templates outside the scanned directory, which correctly stay stubs.

Also

• .twig / .scss / .sass added to _LANG_FAMILY_BY_EXT. With no named family _lang_family returns None, which lets the stub-remap pass bind e.g. an SCSS mixin stub to a same-labeled function in an unrelated language.
• .css is deliberately out of scope. In a Sass project the CSS files are compiled output of the .scss already indexed, so adding it duplicates the stylesheet layer. Worth a separate discussion for hand-written-CSS projects.

Testing

tests/test_twig.py and tests/test_scss.py — 44 new tests, plus classification tests in tests/test_detect.py. Both Twig dialects are covered (Drupal SDC and plain-path/Symfony), so this isn't a Drupal-only patch. Tests pin the behaviors that were actually wrong during development: the docblock self-include, single-line --token: declarations, the target_file stamp, and existence-gating so an unresolvable reference still dangles.

Full suite: 3,757 passed, no new failures. (15 pre-existing failures in test_skillgen.py / test_ollama_retry_cap.py reproduce identically on a clean 0.9.28 checkout, untouched by this branch.)

Known limitation

Design-token nodes don't collapse into one shared node — --spacing-05 referenced from 59 stylesheets becomes 59 per-file nodes, because disambiguation salts them apart like any same-named symbol. Consumers are still findable by label, but defines_token never reaches uses_token across files. The type: "module" anchor exemption (#1327) looks like the right mechanism, but that field doesn't survive the AST node path, so I left it out rather than ship something inert. Happy to follow up if you'd like it fixed.