Fix column in parquet dereference pushdown

[chliang71]

This is to fix this issue.

Test plan - (Please fill in how you tested your changes)

Tested in our internal cluster. Updated unit test

== RELEASE NOTES ==

General Changes
* Fixes a bug in Parquet dereference pushdown. The issue causes column name not picked up properly, resulting in inconsistent query result in certain cases.

[chliang71]

With this change, the check from this patch is failing, so removed for now, will update with new unit test.

[vkorukanti]

LGTM (minor comments).

Please add a test. You can look at HiveLogicalPlanner to construct a specific plan without relying on getting SQL converted to a desired plan.

[vkorukanti]

Should we add check to make sure regularHiveColumnHandles.get(name) returns non-null?

[chliang71]

Added a Preconditions check

[vkorukanti]

Do we need this mapping (columnName in HiveColumnHandle -> HiveColumnHandle)? Expressions should be referring to the name in Assignments. Is that not the case in some cases?

[chliang71]

Sorry are you referring to the second putAll of the two here? If remove this mapping, I would get
nested column [X.Y]'s base column X is not present in table scan output
error in the immediate following logic. This places uses the subfield name to look up back to HiveColumnHandle, and now subfile has the same name as the original HiveColumnHandle, so I used the same map to do this lookup.

[chliang71]

Updated the unit test from this change, I think this test case actually shows the change. Instead of having column name of x_1.a, now it is x_a. cc. @vkorukanti

[chliang71]

@vkorukanti @zhenxiao do you mind helping merge this PR? thanks!