fix(extract): TS bare-path / .svelte.ts / .svelte.js / multi-dot import resolution

[jippi]

Code written by Claude Code, reviewed by me (I'm not super familiar with Python, but SWE with +20 years experience).

Fixes #716. Independent of #714 (#713's PR) — applies cleanly with or without it. The two stack additively for SvelteKit projects.

Problem

_import_js previously only rewrote .js→.ts and .jsx→.tsx, leaving every other common TypeScript / SvelteKit / Vite import shape unresolved. The resulting node id wouldn't match the target file's own _make_id, so build_from_json dropped the edge as external.

Missed shapes — all common in real codebases:

Import	Real file	Was resolving?
import { foo } from './foo' (bare path, TS convention)	foo.ts	no
import { x } from './x.svelte' (Svelte 5 rune file, TS)	x.svelte.ts	no
import { x } from './x.svelte' (Svelte 5 rune file, JS)	x.svelte.js	no
import { x } from './queue' (directory import)	queue/index.ts	no
import { x } from './foo.shared' (multi-dot helper)	foo.shared.ts	no
import { x } from './foo.config'	foo.config.ts	no
import { x } from './foo.integration' (test helper)	foo.integration.ts	no
import { x } from './ambient.d'	ambient.d.ts	no

Fix

A new resolver — _resolve_js_module_path(p) -> Path — mirrors Vite/TypeScript resolver order:

1. exact path (when it's a real file)
2. directory → try index.{ts,tsx,js,jsx}
3. .js→.ts, .jsx→.tsx (existing TS ESM convention, applied before the generic loop so foo.js doesn't match foo.js.ts when foo.ts is the real target)
4. append each candidate extension to the full filename. This subsumes bare paths, Svelte 5 rune files (both .svelte.ts and .svelte.js), multi-dot helper files, ambient declarations, and config files in one rule.
5. directory variant: <name>/index.{ts,tsx,js,jsx}

Falls back to the original path on no match — preserves behaviour for genuinely external modules.

Wired into _import_js (relative + alias branches) and extract_svelte's regex pass for import('...') so static and dynamic imports both benefit.

Subtle: filesystem checks distinguish files from directories (.is_file() / .is_dir() rather than .exists()). Treating any existing path as a file would short-circuit directory imports before the index lookup ever ran.

Priority order note: the _JS_RESOLVE_EXTS tuple puts .ts before .js. This deliberately differs from Vite's resolve.extensions default (which has .js first). graphify is a source-code analysis tool, not a runtime resolver — when a project has both foo.ts and foo.js (typically a .ts source plus a .js build artifact), the source is what we want to graph. Documented in test_resolve_svelte_prefers_svelte_ts_over_svelte_js.

Tests (35 new, all passing)

tests/test_import_extension_resolution.py:

• Resolver unit (19): every branch with positive + negative cases, including TS-vs-JS priority, directory→index priority, .svelte→.svelte.ts and .svelte→.svelte.js rune file variants, hybrid project handling, real .svelte wins over rune-file siblings, multi-dot helper files, ambient .d.ts, partial-name match rejected, dotfile suffix handling.
• End-to-end (10): TS bare-path imports, directory→index, Svelte rune-file imports, type-only import type with bare path, per-symbol named-import edges target the correct id after resolution, alias + directory composes, alias + bare path, multi-dot via tree-sitter pipeline, alias → bare path → rune-file chain, dynamic-import bare path through the regex pass.
• Regression guards (6): explicit .ts/.svelte imports still work, external module specifiers unchanged, real .js not rewritten when .ts absent, dir without index returns input, partial-name rejected, dotfile not treated as bare.

Validation on a real codebase

A 1,873-file SvelteKit app:

State	Edges	vs upstream
upstream (graphifyy v0.7.5)	12,096	—
with this PR	20,151	+67%

Concrete impact:

• Isolated .ts files: 107 → 32 (−70%)
• Of the remaining 32 isolated files: 18 are truly orphan (CLI entry points, root configs, test helpers loaded by name), 11–12 are SvelteKit +page.server.ts / +layout.server.ts files (loaded by the framework's filesystem router, never import-ed), 2–3 are a separate file-collection pipeline bug that I'll file as its own issue.
• A representative .svelte page went from 4 → 7 outgoing import edges, recovering a type-only import, a Svelte 5 rune-file import, and a per-symbol named-import edge.

Test plan

• All resolver unit + edge-case + end-to-end + regression tests pass (35/35)
• Full test suite: 581 pass, 7 pre-existing failures unrelated
• Smoke test on real 1,873-file SvelteKit project
• Manual verification of repaired queries
• Performance: ~9k extra filesystem stat calls per full build, sub-50ms overhead