Improve the creation of ExecutorService when reading footers

Doing some benchmarks loading a large set of parquet files (3000+) from the local FS, we observed some inefficiencies in the number of created threads when reading footers.

By reading, the read the configuration parallelism in Hadoop configuration (defaulted to 5) and allocate 2 ExecuteService with each 5 threads to read footers. This is especially inefficient if there is less Callable to handle than the configured parallelism.

**Reporter**: [Benoit Lacelle](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=blasd)

_{**Note**: *This issue was originally created as [PARQUET-1055](https://issues.apache.org/jira/browse/PARQUET-1055). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}