Use pyspark.InheritableThread as the base thread class for threadpool

[NivekNey]

Context

InheritableThread is a thread class that pyspark provides to support sync between JVM and python threads.

The ThreadPool class used in this package is python's original multiprocessing.pool.ThreadPool.

What goods/bads will it bring? Dunno, I'll start looking around.

Proposal

Without touching this package's source code, a user can already nicely make their own ThreadPool that uses their own Thread class. If this package were to adopt the pyspark Thread class, then the steps may be:

1. Create a custom DummyProcess that inherits pyspark.InheritableThread.
2. Create a custom ThreadPool that uses the custom DummyProcess.
3. Use the custom ThreadPool in this package's SparkDistributedBackend._get_pool method.

[WeichenXu123]

Yes in future we will switch to it. InheritableThread can avoid memory leak in pyspark pin-thread mode (slightly leak not serious).