feat(batch,snapshot): add --runtime workerd

[yusukebe]

`batch` and `snapshot` take `--runtime workerd`, like `request`. One workerd starts from the wrangler config and every step runs in it, so a flow over the real bindings (put to KV, then get; D1; an AI binding) runs in one call with `save` and `expect`. Until now the only way was `request --runtime workerd` per step, which starts workerd each time and cannot carry values between steps. Asked by a user trying an AI-binding middleware.

The workerd code moves to `src/utils/workerd.ts` and splits into a target that can be started once, asked many times, and disposed; `request` keeps using the one-shot form. `runBatch` and `snapshotLines` send to any target with a `request()` method, so a Hono app and a running workerd look the same to them. `snapshot` reads the routes from `main` in-process and sends the requests to workerd. Only `node` and `workerd` are accepted here; `bun`/`deno` point at `request`.

Checked on a real wrangler project: `snapshot --runtime workerd` reports `Cloudflare-Workers` as the user agent, and a KV put-then-get batch passes on one workerd. Known limit, not new: an app that imports `cloudflare:*` modules cannot be imported in-process, so `snapshot` (any runtime) and `routes` fail on it with a build error.

The author should do the following, if applicable

• Add tests
• Run tests
• pnpm run format:fix && pnpm run lint:fix to format the code
• Add TSDoc/JSDoc to document the code