lower segment heap footprint and fix bug with expression type coercion

[clintropolis]

Description

While working on some other stuff I noticed that we were doing something when loading segments which results in some extra heap usage storing redundant stuff we don't really need. Specifically, the changes in #12279 causes us to make a combined column and dimension list in the form of Indexed to pass to the SimpleQueryableIndex, which it was doing prior to this PR with https://github.com/apache/druid/blob/master/processing/src/main/java/org/apache/druid/segment/IndexIO.java#L648 GenericIndexed.fromIterable which is going to make new heap bytebuffers of the column and dimension lists. The columns Indexed is thrown away basically, but the dimensions list is kept around forever.

We actually already have to have all of these column names on heap, and are already interning them via the FileSmooshMapper https://github.com/apache/druid/blob/master/processing/src/main/java/org/apache/druid/java/util/common/io/smoosh/SmooshedFileMapper.java#L88 which does so whenever it reads meta.smoosh to get the internal file list of a smoosh. There is an internal file as the entry point for every column name, so if we re-use this interner for the map of ColumnHolder suppliers as well as to back the Indexed<String> for the dimension list then we can save some heap.

With my laptop test cluster of ~5k segments it reduces the heap footprint by ~20MB

Before:

After:

Looking specifically at strings, we now have about 1/3 as many on heap (from the map of ColumnHolder suppliers), not counting all of the heap bytebuffers which were backing the dimensions lists.

Completely unrelated, this PR also fixes a bug with native expression type coercion with complex types. The added test would fail where it would return "unknown complex" sometimes when using the ExpressionTypeCoercion.operator method. I guess I could have made it its own PR, but both things were small...

This PR has:

• been self-reviewed.
  • using the concurrency checklist (Remove this item if the PR doesn't have any relation to concurrency.)
• added Javadocs for most classes and all non-trivial methods. Linked related entities via Javadoc links.
• added comments explaining the "why" and the intent of the code wherever would not be obvious for an unfamiliar reader.
• added unit tests or modified existing tests to cover new code paths, ensuring the threshold for code coverage is met.
• been tested in a test Druid cluster.

[imply-cheddar]

typo: s/very/every/