feat(extraction): add Slint language support

[ZyphrZero]

Summary

• Add Slint (.slint) as a supported language with a vendored ABI-15 tree-sitter grammar.
• Extract Slint components, globals, interfaces, structs, enums, properties, callbacks, functions, imports, child component references, and call edges.
• Fix Slint re-export barrels such as export { AboutPage } from "about_page.slint"; so barrel-only files create import nodes and resolved cross-file imports edges.
• Document Slint support in the README, docs language reference, changelog, and add Slint repos to the reusable agent-eval corpus.

Refs #648
Refs #916

Validation

• npm run build with Node v24.14.0: passed.
• npx vitest run __tests__/slint-extraction.test.ts: passed, 5 tests.
• npx vitest run __tests__/slint-extraction.test.ts --pool=forks --maxWorkers=1 --minWorkers=1: passed, 5 tests.
• npx vitest run __tests__/extraction.test.ts -t "React|TSX|component": passed, 14 selected tests.
• CLI smoke: indexed a temporary .slint project with dist/bin/codegraph.js; status reported languages: ["slint"], 1 file, 6 nodes, 8 edges.
• Re-export smoke: indexed a temporary pages.slint barrel and verified SQLite edges from pages.slint to AboutPage, TableViewPage, and TableViewPageAdapter in the target files.

Full-suite status on Windows

• npm test is not green in this Windows checkout. The failures are outside the Slint extractor path: temp-dir EPERM cleanup in JVM/C++/PHP/MCP tests, MCP initialize/roots Windows file-lock failures, worktree tests comparing short ADMINI~1 paths with long Administrator paths, and a Vitest worker OOM when running the large extraction.test.ts file.
• I did not change non-Slint test infrastructure in this PR; the Slint-specific tests and focused regression checks above pass.

Add-language extraction benchmark

Ran scripts/add-lang/bench.sh slint <repo> <url> "<question>" headless with the local PR build via CG_BIN=/tmp/codegraph-dev-bin/codegraph on 2026-06-30. After the re-export fix, extraction was rerun and the large repo index grew as expected.

Tier	Repo	Indexed files	Nodes	Edges	Extraction
Small	slint-nodejs-template	2	9	10	PASS
Medium	SurrealismUI	244	3,439	3,744	PASS
Large	slint	2,694	42,001	143,353	PASS

Paid agent A/B benchmark

Headless Claude Code A/B (--model sonnet --effort high, --max-budget-usd 4) comparing local CodeGraph MCP vs no CodeGraph MCP.

Repo	Arm	Tool calls	Read	Bash/Grep	CodeGraph calls	Duration	Cost
slint-nodejs-template	with CodeGraph	2	0	0	2	17s	$0.194
slint-nodejs-template	without CodeGraph	5	3	2	0	17s	$0.189
SurrealismUI	with CodeGraph	5	0	0	5	55s	$0.426
SurrealismUI	without CodeGraph	19	15	4	0	71s	$0.471

Previous valid large-repo paid A/B run, before the re-export fix:

Repo	Arm	Tool calls	Read	Bash/Grep	CodeGraph calls	Duration	Cost
slint	with CodeGraph	15	7	1	7	75s	$0.765
slint	without CodeGraph	20	12	5	0	86s	$0.562

The large-repo paid A/B rerun after the re-export fix was blocked by API 402 Insufficient Balance in both arms, so I am not claiming post-fix large A/B numbers. The post-fix extraction benchmark did complete and shows the larger slint repo index increasing from 41,774 nodes / 142,923 edges to 42,001 nodes / 143,353 edges.

Notes

• The benchmark exposed a real Slint support gap: barrel files like examples/gallery/ui/pages/pages.slint only re-export symbols and previously indexed with no useful dependencies. This PR now links those barrels to the exported target symbols.
• The small and medium Slint repos reached 0 Read/Bash with CodeGraph in the latest paid rerun. Medium still improved duration/cost; small had fewer tool calls and no reads, but similar duration and slightly higher cost in that rerun.
• The previous valid large official slint run reduced duration, total tool calls, Reads, and Bash/Grep usage, but still fell back to 7 Reads and cost more than the no-CodeGraph arm. The residual fallback is around the mixed Slint/Rust include_modules! / generated-runtime boundary, so this PR proves useful Slint extraction but does not claim complete cross-language macro/runtime flow coverage.