Drop two-phase command-line parsing

[Eyal Rozenberg]

We currently parse our command-line in two phases: One kernel-inspecific, then another that's kernel-specific, where the arguments and required preprocessor definitions can be passed as command-line arguments.

Benefits:

1. We get a necessary argument and define listing "for free"
2. Can keep the initial parsing entirely independent of which kernel we want to run

Detriments:

1. More code.
2. More complex, non-C++-idiomatic code.
3. Kernel parameter names can clash with basic command-line argument names
4. One-char kernel parameters need to be handled differently than multi-char.

Now, actually, in hindsight, we can achieve benefit (2.) and take the kernel arguments, by passing --argument name=val 's. So, let's drop the existing scheme and just do that instead.