Rethink composition of experiments and value precedence

[smarr]

At the moment, we compose experiments/runIs bottom up, with the highest elements having priority. See https://rebench.readthedocs.io/en/latest/config/#priority-of-configuration-elements

Though, in some cases this doesn't seem to be very useful.

For instance, if I want to override the number of iterations that a benchmark is supposed to run, I can't, if it's a per-benchmark setting, which can be useful for large, diverse benchmark suites.

So, if I want to limit the number of iterations for a specific experiment, I need a separate suite definition, as I have been doing for many of my configurations.

Though, things might be more reusable, if priority would work the other way around.

Or, if I could mark a specific value as important.
Perhaps similar to https://developer.mozilla.org/en-US/docs/Web/CSS/Specificity#the_!important_exception

[smarr]

A first version is implemented with #170, but it's only partial.