fast import: tenant broken if import cancelled / errors out before /status/pgdata reaches done state

[problame]

If the import flow stops due to an error or cancellation (PS restart, tenant migration), the index part in S3 has an invalid disk_consistent_lsn and the tenant will fail to load_remote_timelines in any location where it subsequently attached

Timeline 888a599bcfba087241276c6df8efc311/a6aedf56204960c7b5b9fdc857dff15c has invalid disk_consistent_lsn

More context: https://neondb.slack.com/archives/C033RQ5SPDH/p1734549772067199?thread_ts=1734368383.258759&cid=C033RQ5SPDH

When fixing this issue, also think about / add regression test for resilience to tenant migrations, etc in other places.
Probably there will be no time to get the kind of excellent test coverage we have for the timeline deletion flow, but think about it in the same way.

Evidence

Index part looks like this

cs@cs-neon-mbp:[~/Desktop]: aws s3 cp s3://neon-dev-storage-us-east-2/pageserver/v1/tenants/888a599bcfba087241276c6df8efc311/timelines/a6aedf56204960c7b5b9fdc857dff15c/index_part.json-00000001 - | jq
{
  "version": 10,
  "import_pgdata": {
    "V1": {
      "InProgress": {
        "idempotency_key": "todo",
        "location": {
          "AwsS3": {
            "region": "us-east-2",
            "bucket": "neon-dev-bulk-import-us-east-2",
            "key": "import-pgdata/fast-import/v1/br-little-lab-w28y1bqu"
          }
        },
        "started_at": "2024-12-18T15:33:22.045125540"
      }
    }
  },
  "layer_metadata": {},
  "disk_consistent_lsn": "0/0",
  "metadata_bytes": {
    "disk_consistent_lsn": "0/0",
    "prev_record_lsn": null,
    "ancestor_timeline": null,
    "ancestor_lsn": "0/0",
    "latest_gc_cutoff_lsn": "0/0",
    "initdb_lsn": "0/0",
    "pg_version": 15
  },
  "lineage": {}
}