Incrementally complete the packaging plan

[axch]

PR #101 leaves several items from the (ambitious) plan of Issue #95 undone. We should either do, abandon, or officially defer (perhaps in a separate ticket) each of the following items:

• [ ] The actual command line program $ aronnax simulate [OPTIONS] [directory]. This is a straightforward exercise in parsing command line arguments and mapping them to options to driver.simulate.
• [ ] NetCDF input, including interpolation to cover the desired placement of grid points.
• [ ] In-memory numpy array input (use as-is should be a very easy change to core.interpret_requested_data; perhaps interpolation may be called for?).
• [x] A full suite of generators for custom idealized input (currently only supports depths and meridional wind, since that's all the examples called for, and I'm not sure what grid configuration to use for which of the others).
• [ ] NetCDF output (Issue #30).
• [ ] Control of the input and output directories, in particular potentially to places outside the "working directory".
  • Is this actually necessary?
• [ ] Accepting arbitrary (presumably postfix) compiler options, and recompiling the core therewith as needed.
  • Implementing this while avoiding redundant recompilations even if options are given seems to fairly naturally allow removing the dependency on make even for the "standard" option sets
  • Note: Aronnax may not be able to write its installation directory after installation, because it may have been installed with administrative privileges but run without. Therefore, the working directory is a reasonable place to put the compiled core executable.
• [ ] Accepting arbitrary prefix modifiers to the core execution command, not just the two presets for valgrind and perf.
  • Also, the perf preset should not include my personal CPU's non-portable instruction mask for floating point operations.