Fix scan crash when framework cache hits persisted review_cache

[MacHatter1]

Problem

desloppify scan could crash while persisting state if state["review_cache"] contains framework detection / info dataclass instances (non-JSON-serializable).

Fix

• Store JSON-safe dict payloads in review_cache for:
  • frameworks.ecosystem.present:* (framework detection)
  • framework.nextjs.info:* (Next.js info)
  • Decode these payloads back to the expected runtime dataclasses when read.
• Harden json_default to encode dataclasses/enums defensively so future cache regressions don't brick scans.

Tests

• pytest -q (6236 passed, 165 skipped)
• Manual scans using workspace code via PYTHONPATH:
  • vercel/nextjs-postgres-auth-starter
  • vercel/swr
  • sindresorhus/ky

[peteromallet]

Thanks @MacHatter1! You correctly identified the framework cache serialization crash — that's a real bug. The json_default dataclass handler part of your PR was implemented independently (commit 61bb7cb on 0.9.11), and the root cause was separately fixed by PR #483 (commit 986f884 by @maciej-trebacz) which introduces a runtime_cache to keep framework objects out of persisted state entirely.

Your encode/decode helpers for framework cache round-tripping addressed a secondary issue (cache miss after reload), but since PR #483's approach means framework objects never reach the persistence layer, that code path is no longer needed.

Appreciate the thorough investigation — you clearly dug deep into the serialization chain!