[v3 Enterprise] PachaTree engine: tables without tags return 0 rows although the data is persisted

Steps to reproduce:

1. Run InfluxDB 3 Enterprise 3.11.4 on the upgraded storage engine (--upgrade-pacha-tree, single node, --mode all, --object-store file).
2. Write a table without tags and a control table with a tag via /api/v3/write_lp:
   复制

   notag value=1.0 <ts>            (7 lines)
   withtag,host=a value=1.0 <ts>   (7 lines)

3. Query immediately: SELECT count(*) FROM notag → 7, SELECT count(*) FROM withtag → 7.
4. Wait ~60 s for the snapshot. SELECT * FROM system.pt_ingest_files now lists the gen0 file for notag with row_count = 7 and series_key: SeriesKey::V2 { tags: [] }.
5. Query again: SELECT count(*) FROM notag → 0, withtag → 7. A restart does not change this.

Same effect for migrated data: after a Parquet → PachaTree upgrade (Cluster upgrade complete, all system.upgrade_parquet sources compacted, no errors logged), six tag-less tables with ~43k rows each return 0 rows, while all 1,100+ tables with tags match their pre-upgrade row counts exactly.

Expected behaviour: Tables without tag columns are queryable on the PachaTree engine, as they are on the Parquet engine (control: same write + snapshot + restart on Parquet 3.11.4 → rows remain). Alternatively, writes to tag-less tables are rejected with a clear error (the CLI already requires --tags for create table).

Actual behaviour: All queries against a tag-less table (SELECT *, count(*), time-range filter, field aggregate, InfluxQL) return no rows as soon as the data is in .pt files. The data itself is persisted:

• influxdb3 export data --database <db> --table notag exports the compacted windows; read with pyarrow they contain exactly the pre-upgrade row count and time range.
• EXPLAIN SELECT count(*) FROM notag shows normal PachaTreeWindowExec … run_sets=[…] est_rows=… nodes, identical in shape to a tag table.

Environment info:

• Official release binary influxdb3-enterprise-3.11.4_linux_amd64.tar.gz (revision 6ee8b0e752), not built from source.
• Linux 7.0.14-15-pve x86_64, Debian 13 in an unprivileged LXC container on Proxmox VE.
• 4 GB RAM, 4 vCPU (2 licensed cores, Home license), local ZFS-backed disk, --object-store file.

Config:

influxdb3 serve --node-id primary-node --mode all --cluster-id <id> --object-store file --data-dir <dir> --license-type home --num-cores 2 --upgrade-pacha-tree --tls-cert <cert> --tls-key <key>

All other options default. Reproduced identically with additional Parquet-era options (--compactor-input-size-budget, --wal-snapshot-size, --exec-mem-pool-bytes, gen1/snapshot env vars).

Logs: No errors or warnings related to the affected tables; startup and migration logs are clean. Possibly related earlier fixes: 3.9.2 "Fixed compaction for tables with no tags (empty series key)", 3.11.3 "Missing rows from an OR predicate against a file index".