Fix handling of input size configuration and respect default_experiment setting

[IsaacOscar]

This minor change allows variables defined for individual benchmarks to override those in the benchmark_suites. In particular, this allows individual benchmarks to have different input_sizes:

benchmark_suites:
    my-suite:
        command: "harness --input-size=%(input)s %(benchmark)s"
        benchmarks:
            - my-benchmark1: { input_sizes: [1, 2] }
            - my-benchmark2: { input_sizes: [3, 4] }

Running my-suite will now run:

my-executor harness -input-size=1 my-benchmark1
my-executor harness -input-size=2 my-benchmark1
my-executor harness -input-size=3 my-benchmark2
my-executor harness -input-size=4 my-benchmark2

Previousesly, it would run:

my-executor harness -input-size=None my-benchmark1
my-executor harness -input-size=None my-benchmark2

[coveralls]

Coverage increased (+0.08%) to 85.881% when pulling e5a90c4 on IsaacOscar:master into b75a2a0 on smarr:master.

[IsaacOscar]

I have also added a fix to respect the "default_experiment" key if it is present in the yml file. Previously, it was ignored.

[smarr]

Hi @IsaacOscar thanks a lot for the fix.
This is indeed a bug and should have already worked. The schema explicitly allows it here: https://github.com/smarr/ReBench/blob/master/rebench/rebench-schema.yml#L116

The EXP_VARIABLES are the three things defined here: https://github.com/smarr/ReBench/blob/master/rebench/rebench-schema.yml#L81-L101

I'll likely not get to it myself this week, so, I was wondering, would it be possible to add tests for these things?
I had a somewhat similar issue earlier, and added tests for it here: https://github.com/smarr/ReBench/pull/113/files#diff-b1e4aefa3d9f06036f4772771d13c2e5

[smarr]

I added some tests, and the PR seems ready to merge now.

Thanks a lot Isaac!