A more realistic allocations profile

[ivan-tkatchev]

The even/linear/exponential distribution of allocation sizes is a good stress test but doesn't correspond to real-world usages of malloc.

Could you add one more test case with a distribution sampled from real workloads?

Here's an example of allocations when gcc is compiling lockfree-malloc:
allocs_sample.tsv.txt

(Having several test cases for different real-world workloads would be even better, I guess.)

Thank you.