Using real database instead of using `.json` files as database

Advanced of #1894

I found a performance bottleneck: `atomicWrite` of `sites.json`.
This is significant when seeding many sites + sites frequency update.

When using `.json` files as database, if one row updates, whole file need to rewrite.
When using real database, if one row updates, only related disk sector need to rewrite.

The issue names only atomicWrite and sites.json, with no repository files or tests. Start by locating that write path and the seeding and frequency-update callers; clarify the database choice, migration scope, and measurable completion criteria before implementation.

索引モデルが issue の本文から書いたものです。