Kernel extraction should support MPI, and potentially several different files per kernel

[hiker]

The restriction on not supporting distributed memory should be easy to lift, by just adding a rank number to the output files. The driver would then not read only one, but several files (one per process) and execute the kernel. Once we have this, we might also want to create more than one file per kernel (per process). For example maybe the first and last time that the kernel is called, or store X data files somehow equally distributed across time, ... Again, this would add a similar number to the output files to the rank, causing similar issues in the driver as outlines below.

• [ ] Enable distributed memory support for kernel extraction by adding a rank number to the output files.
  • [ ] Make the driver either accept a rank number to read the right file, or automatically read the file for all ranks and execute the kernel. Also think of #1989 , how to present this information in a short summary.
• [ ] Enable that a kernel can dump files more than once, again by adding an index number (and similar to the MPI number let the driver handle this). We might use techniques from managing backup tapes to decide which files to delete to keep a kind of 'equidistant' set of files. Maybe an overall max size/kernel could be specified, or a number/kernel??