Add NoSQLExtractor for raw MongoDB driver usage #76

Context
The EXTRACTORS chain in extractors/index.ts currently covers two schema sources: PrismaExtractor (full field/relation data) and SQLExtractor (raw pg/mysql2/better-sqlite3 clients → table names as pseudo-entities, no field data). Projects using the native mongodb driver without an ODM have no coverage and fall through to the low-confidence route-hint fallback.

Scope
New NoSQLExtractor implements IEntityExtractor, same pattern as SQLExtractor:

• canHandle(): detect mongodb import (content-based, same regex approach as usesRawSqlClient)
• extract(): scan .collection("name") / .collection('name') calls via ts-morph AST string literals (not regex-only — same rationale as SQLExtractor's comment on quote-pairing failure modes)
• Output: pseudo-entities, fields: [], relations: [], source: "nosql" — NoSQL has no schema metadata to read the way Prisma does
  Acceptance criteria
• Detects raw mongodb driver usage only (Mongoose is a separate extractor, see Render AI markdown cleanly in the terminal #4)
• Registered in EXTRACTORS, after the ORM extractors, alongside/after SQLExtractor
• EntityInfo["source"] / EntityGraph["source"] union types in types.ts extended with "nosql"
• Tests added mirroring entity-extractors.test.ts coverage for SQLExtractor