Avoid relying on barrier_interval_ms to increase operator cache eviction watermark

[hzxa21]

After #16087, we have switched to using sequence-based memory eviction policy instead of an epoch-based policy. This means we no longer need to rely on barrier to increase the watermark. However, it seems that we are still use barrier_interval_ms as the ticker interval here. This may result in delay in cache eviction when barrier_interval_ms is increased.

[yuhao-su]

This is a long existing issue. We should first make sure all executors are not evicting cache on barrier basis.

[hzxa21]

This is a long existing issue. We should first make sure all executors are not evicting cache on barrier basis.

I think we already did? Just checked hash join and hash agg.

[MrCroxx]

This is a long existing issue. We should first make sure all executors are not evicting cache on barrier basis.

It had been done before the sequence-based operator cache eviction.