Discussion around storing SUM as a optional metadata field.

### Describe the enhancement requested

Currently parquet only stores fields for predicate pushdown for scan pruning, but we can also keep `sum` in metadata field, while this won't help in predicate pushdown, but it can act as a pre-aggregation for sum and avg queries.

While this may be challenging and bit complex because we would need to handle sum/overflow etc. But this can be optional just like bloom filters and would be helpful for many usecases. And shouldn't have much of a overhead due to it being a stateless operation.

Start by reviewing the existing Parquet metadata fields and optional bloom-filter handling described in the issue. Define how optional SUM metadata would be represented and how overflow and SUM/AVG pre-aggregation should work. Done means the format design and its trade-offs are agreed.

Written by the indexing model from the issue text.