feat(al): add AL (Business Central / Dynamics 365) language support

[Capitaller]

Summary

Adds full AL (Microsoft Dynamics 365 Business Central) language support to CodeGraph, enabling semantic code intelligence across large ERP codebases.

What's included:

• New alExtractor in src/extraction/languages/al.ts with a resolveName hook that correctly resolves AL object names from the object_name field (Codeunit, Table, Page, Report, XMLport, Query, and all extension types)
• .al extension registered in src/extraction/grammars.ts
• 'al' language type added to src/types.ts
• Case-insensitive reference resolution in src/resolution/name-matcher.ts (AL is case-insensitive by language spec)
• Extraction parity test in __tests__/al-extraction.test.ts
• Uses the tree-sitter-al WASM grammar (bundled under src/extraction/wasm/)

Motivation

AL is the primary language for Microsoft Dynamics 365 Business Central, one of the most widely deployed ERP platforms globally. A typical BC codebase (e.g. the official W1 BaseApp) contains 7,800+ files, 115,000+ symbols and 264,000+ dependency edges.

Without CodeGraph, AI agents are forced to rely on grep/find, which fails to understand the graph of Codeunit dependencies and produces shallow, keyword-matched answers.

With CodeGraph, an AI agent answers this in 47s using 2 tool calls:

"Which Codeunit and function is responsible for calculating invoice rounding when there is a 100% prepayment on a Sales line? What temporary tables does it use?"

The agent correctly traces the full call chain: GetSalesLines → InsertInvoiceRounding → InitInvoiceRoundingLine, identifies the responsible Codeunit (Sales-Post Prepayments, Codeunit 442), and names the correct temporary tables (TempPrepmtInvLineBuf, TempSalesLine, TempVATAmountLine). Without CodeGraph, the same agent runs 10+ grep commands and produces a shallow answer that misses the core architecture.

Test plan

• npm run build — passes with zero TypeScript errors
• npm run test __tests__/al-extraction.test.ts — AL extraction parity test passes (Codeunit name, procedure nodes, and call reference edges all verified)
• npm run test — full suite passes: 175 test files, 3,010 tests, 0 failures
• Indexed 7,820 files from the official Microsoft BC W1 BaseApp source → 115,609 nodes, 264,588 edges in 11.3s
• Verified AI agent uses codegraph_explore MCP tool calls exclusively — no grep/find fallback — on real Business Central architecture queries