Enhance TaskContext and add task failure root cause

**Is your feature request related to a problem or challenge? Please describe what you are trying to do.**
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
(This section helps Arrow developers understand the context and *why* for this feature, in addition to the *what*)

In current Datafusion/Ballista plan's execution path, the error result mapping is a chaos, the root cause of the task failure is not clear, especially when the plan tree contains pipeline breakers.
To make the error handling clear and make it easy to reasoning about the real task failure reason, the suggestion is to add a new method to TaskContext to set the task failure root cause into TaskContext.

Below is an example of Ballista shuffle read execution path, the arrow '->' represents the error/result return flow.

`
BallistaClient.fetch_partition(): BallistaError::General -> Ballista ShuffleReaderExec.fetch_partition(): DataFusionError::Execution -> Ballista ShuffleReaderExec.execute() : SendableRecordBatchStream ->
->RecordBatchStreamAdapter.poll_next(): ArrowResult::ExternalError -> `

We can see the root error is mapped to Ballista Error -> DataFusion Error -> Arrow Error. And depends on how the
stream is polled and consumed, the Arrow Error can be mapped to DataFusion Error or Ballista Error again.
There are back and forth error mapping in the code base, this makes the root case reasoning become a challenge.

Want to hear your thoughts.

@andygrove @alamb @thinkharderdev

**Describe the solution you'd like**
A clear and concise description of what you want to happen.

**Describe alternatives you've considered**
A clear and concise description of any alternative solutions or features you've considered.

**Additional context**
Add any other context or screenshots about the feature request here.

Start with TaskContext and trace the Ballista shuffle read path described in the issue: fetch_partition, ShuffleReaderExec.execute, and RecordBatchStreamAdapter.poll_next. Map where Ballista, DataFusion, and Arrow errors are converted, then determine the TaskContext API and behavior needed to preserve the task failure root cause. Done when the intended root cause remains clear through the execution and stream-polling path, with tests covering the mappings.

Written by the indexing model from the issue text.