Infer benchmark data to use from keyword arguments

[eriknw]

Similar to #6, but has a different set of challenges. To summarize, I would like to be able to provide values to use for benchmarks via keyword arguments:

def bench(data=[0, 10, 10000]):
    zeros(data)

When displaying output, benchtoolz should be intelligent enough to use the name of the input if it is given, such as:

data_empty = 0
data_small = 10
data_large = 10000

def bench(data=[data_empty, data_small, data_large]):
    zeros(data)

Another way to label input may be to allow a dictionary to be given:

def bench(data={'empty': 0, 'small': 10, 'large': 10000}):
    zeros(data)

I don't know if a dictionary should be supported, because best practice will be to use keyword arguments for simple data, such as someflag=[True, False].

Another variant (pitfall?) to be aware of is if values are pre-initialized in a list, such as:

bench_data = [0, 10, 1000]

def bench(data=bench_data):
    zeros(data)