Benchmark Suite

[dannys4]

Since this is more-or-less "functional" (withholding lingering issues from #46 ), many of the performance improvements will be incremental. It would be beneficial to have some suite of tests demonstrating benchmarks that make it easier to understand how helpful a performance change is and compare across changes. Some points it should make:

• Power of two, three, and four scaling
• Power of two/four * three
• Power of two/four * small prime number
• Power of two/four * "large" prime number
• "large" prime number
• "large" prime * "large" prime
• the product of many unique small primes (e.g., 3571317)
• one or many of the above with "eccentric" types (symbolics, static/structured arrays, forward-diff, etc)

This list is not exhaustive. It would make sense that any performance PR should improve at least one of these and shouldn't decrease any of them (or at least, shouldn't make them more than ~1% worse, which could happen depending on function overhead etc). Could be a new repo, part of the tests, but likely should just be a small package extension for now (I don't think the intention here is to provide de facto benchmarking for all AbstractFFTs, just a useful suite to understand the effect of PRs). It shouldn't be part of the main package to mitigate extra dependencies, extraneous compilation, etc.

I'll try to take a stab at this when I get some time.