[FEA] Plugable Cache - Githubissues

This issue deals with creating a plug-able cache and will be changed as we find more stuff that needs to be worked on.

[x] Create a Plug-able cache serializer assuming we have the plugin enabled and the data is all going back on the host.
[x] InMemoryTableScanExec needs to be implemented which needs to be smart enough to make sure we don't copy the data back to host unless its needed.
[x] Performance analysis. This will mostly be testing how fast are we reading the cache back compared to the CPU.
[x] Come up with a real world example. This was done as a part of the performance analysis, nothing was checked-in yet.
[x] We see a lot of row/col conversions, get rid of the unnecessary ones.
[x] Reading CachedBatch to InternalRow
[x] Write InternalRow to CachedBatch
[x] Write ColumnarBatch to CachedBatch on CPU
[x] Read ColumnarBatch from CachedBatch on CPU
[x] Handle CalendarIntervalType and NullType in Cached Batch. Issue/PR
[x] Handle nested types with CalendarIntervalType and NullType. Issue/PR
[x] Improve the Parquet Serializer to handle large batches that may exceed Int.MAX_SIZE Issue/PR

NVIDIA / spark-rapids