Fixes #2083

The `queryMinRefId` query uses
```sql
SELECT id FROM <table> WHERE <field> >= $1 ORDER BY id ASC LIMIT 1.
```

The planner sometimes picks a bad plan:

```sql
Index Scan using tx_pkey on tx
  Filter: (block_id >= $1)
```

the filter is not Index Cond, so this ends up in a sequential scan.
The index refers to the primary key and is only used for sorting.

Instead we use a simpler query without ORDER BY:

SELECT id FROM <table> WHERE <field> >= $1 LIMIT 10000
This forces the planner to use the field's index.
The results are fetched and the minimum is found in Haskell.
Near the tip this returns only a handful of rows.

If there are more than 10000 matching rows (large rollback),
we fall back to the original ORDER BY id ASC LIMIT 1 query.