arrow

ARROW-9452: [Rust] [DataFusion] Optimize ParquetScanExec

This PR changes the way channels are used to communicate with the parquet scan thread. Rather than the caller send a request per batch, the thread just pushes batches to a bounded channel. This is far more efficient as shown by the almost 2x improvement in the benchmark crate.

Before:

```
Running benchmarks with the following options: Opt { debug: false, iterations: 3, batch_size: 4096, path: "/mnt/nyctaxi/parquet/year=2019", file_format: "parquet" }
Executing 'fare_amt_by_passenger'
Query 'fare_amt_by_passenger' iteration 0 took 2267 ms
Query 'fare_amt_by_passenger' iteration 1 took 2221 ms
Query 'fare_amt_by_passenger' iteration 2 took 2224 ms
```

After:

```
Running benchmarks with the following options: Opt { debug: false, iterations: 3, batch_size: 4096, path: "/mnt/nyctaxi/parquet/year=2019", file_format: "parquet" }
Executing 'fare_amt_by_passenger'
Query 'fare_amt_by_passenger' iteration 0 took 1233 ms
Query 'fare_amt_by_passenger' iteration 1 took 1203 ms
Query 'fare_amt_by_passenger' iteration 2 took 1221 ms
```

Closes #7743 from andygrove/ARROW-9452

Authored-by: Andy Grove <andygrove73@gmail.com>
Signed-off-by: Wes McKinney <wesm@apache.org>