#2371 s3 datastore: first run of a new model writes a monolithic .datastore-index.json into a shard-first (v2) bucket

Found while reproducing swamp-club#2353 on localstack (swamp 20260922, @swamp/s3-datastore 2026.09.22.2).

Problem

The bucket was shard-first: _index/_meta.json version 2. The first method run of a brand-new model logged pullIndex.discover with n=86 and wrote a monolithic repo/.datastore-index.json object to the bucket. v2 repos are meant to skip the monolithic index entirely; the shards under _index/ are the source of truth.

Why it matters

A stale monolith in a v2 bucket can be picked up by code paths or older clients that fall back to the monolithic index (pullIndex), giving them a partial or outdated view. It also costs an extra full-index upload on the discovery path.

To reproduce

1. Set up an s3 datastore on an empty bucket so it initialises as v2.
2. Create a new model on one machine and run a method.
3. List the bucket prefix: .datastore-index.json appears alongside _index/.

Expected

On a v2 bucket, discovery and first-run paths never write the monolithic index.