INCREMENTAL_BY_UNIQUE_KEY models not taking batch_size into account

[erindru]

On Trino/Iceberg (and potentially other engines; unconfirmed), the following model doesnt backfill correctly:

MODEL (
    name  datalake_iceberg.stg_premium_day4,
    kind INCREMENTAL_BY_UNIQUE_KEY (
         unique_key id,
         batch_size 1
    ),
    start '2020-01-01',
    cron '@daily'
);

select * from datalake_iceberg.seed_model --note: this is the seed_model from `sqlmesh init`
where event_date between @start_dt and @end_dt

The goal here is to backfill a "daily" model, in batches of 1 day.

However, it does not execute correctly. It triggers a DELETE+INSERT (where the DELETE has WHERE TRUE and clears the whole table) for every interval/batch, even though it should only trigger a DELETE+INSERT for the first batch (to support restatements) and then a MERGE for subsequent batches.

Killing sqlmesh plan before it finishes with ctrl+c and then running sqlmesh plan again, which picks up where it left off, triggers the correct MERGE behaviour.

It appears that the batch_size is not being taken into account correctly when deciding whether to clear the table or merge into the table