Common utility code for dumping to a distributed filesystem

[jlowe]

Is your feature request related to a problem? Please describe.
There are a number of places in the code where data can be side-band written to a distributed filesystem, e.g.: file scan input dumps, GPU core dumps, profiling data, etc. A lot of this code has to solve the problem of how to get the proper Hadoop configuration and credentials on the executor to access the filesystem to create the new file, and that code is often replicated across subsystems.

Describe the solution you'd like
There should be a common API in the executor plugin that subsystems on the executor can use to create a file on the distributed filesystem. The API should allow specifying a specific filename or generating a temporary name based on a filename template.