Parquet reader thread errors do not make query fail

[andygrove]

Describe the bug
The Parquet reader thread can terminate due to errors but it does not make the query fail. It just uses println! to print the error.

Parquet reader thread terminated due to error: Execution("Failed to map column projection for field l_orderkey. Incompatible data types Int32 and Int64") for files: [PartitionedFile { file_meta: FileMeta { sized_file: SizedFile { path: "/mnt/data/tpch/sf100-parquet//lineitem/part-13.parquet", size: 1165760107 }, last_modified: Some(2022-02-05T18:23:00.416225870Z) }, partition_values: [] }]
Query 3 iteration 1 took 5452.6 ms and returned 0 rows

To Reproduce
Run one of the unsupported tpch queries or run against invalid parquet files.

Expected behavior
Query should fail.

Additional context
None

[alamb]

I think this comes from https://github.com/apache/arrow-datafusion/blob/master/datafusion/src/physical_plan/file_format/parquet.rs#L230

We can probably adapt a similar approach as taken in RepartitionExec https://github.com/apache/arrow-datafusion/blob/master/datafusion/src/physical_plan/repartition.rs#L412-L449 with an extra task that handles a panic (join handle returning null) and sends a Error as @andygrove suggests

[alamb]

I'll take a stab at this

Also, as noted by @Dandandan / @andygrove in the slack channel, the #1617 may make this redundant anyways