[Coral-Schema] Fix incorrect type derivation for repeated field reference on UDF calls

[KevinGe00]

What changes are proposed in this pull request, and why are they necessary?

This is a continuation of #507. #507 reverts some of the effects of #409 (namely, around single union type handling), and this PR reverts more effects (see below). Note this PR includes those same reverts.

Incorrect behavior after reverting #409 changes:
In Hive,

Table t1 that has the columns:
struct_col           	struct<union_field:uniontype<string>>

View v1: select extract_union(struct_col).union_field.tag_0 as b from t1
b: string
Schema for col b produced by coral-schema represents:
struct<tag_0:string>

In the above scenario, we expect that produced schema (by coral-schema) for col b to be simply a string, however, what we get back is of type struct<tag_0:string>.

Although #409 mitigates this issue by detecting and unwrapping structs with field tag_0, it doesn't address the root cause. Which is a bug in SchemaRexShuttle.visitFieldAccess that doesn't handle nested field references on UDF calls introduced in #203, and is fixed in this PR here.

How was this patch tested?

Unit tests
Regression tests

[1fanwang]

Mostly LGTM, minor comments

[ljfgem]

Overall LGTM.
Please ensure the regression tests pass for coral-trino, coral-spark and coral-schema.