Skip Computing Nulls for Non-Nullable Parquet Columns

**Is your feature request related to a problem or challenge? Please describe what you are trying to do.**

Currently `DefinitionLevelDecoder` always computes a null mask if any parent of a column is nullable.

This is not only wasteful, but is also dubiously correct as it can result in StructArray that would fail the nullability checks:

Consider the case of

```
optional group l1 {
optional group l2 {
required INT32 leaf;
}
}
```

PrimitiveArrayDecoder will decode leaf with a null buffer, `l2` will then be decoded as a `StructArray` without a null buffer, with leaf as a child. This is technically ill-formed as the Field for leaf will state it isn't nullable, however, leaf has a null buffer

**Describe the solution you'd like**

`DefinitionLevelDecoder` should skip decoding a null buffer for non-nullable columns, some additional care may be necessary for dictionaries to ensure that the dictionary offsets are valid in isolation.

**Describe alternatives you've considered**

**Additional context**

Locate DefinitionLevelDecoder and trace how it creates null buffers for nested optional and required Parquet columns, including the PrimitiveArrayDecoder and StructArray path. Check the dictionary case mentioned in the issue; done means non-nullable columns no longer receive unnecessary null buffers while dictionary offsets remain valid and the nested example is well-formed.

Written by the indexing model from the issue text.