adds a benchfella benchmark

[mindreframer]

problem:

• in PR Use custom underscore function instead of Macro.undescore #5 I wanted to compare different implementations for underscoring

solution:

• use benchfella package
• small script

results:

• the Macro.underscore version is roughly 2x faster... that time adds up on larger json payloads, so I'll keep it for now. It's always possible to switch it to something custom later, if that becomes a problem.

## BasicBench
macro_underscore - short      100000   11.23 µs/op
regex_underscore - short      100000   26.51 µs/op
macro_underscore - long        50000   45.81 µs/op
regex_underscore - long        20000   75.77 µs/op

[mindreframer]

@anronin finally came to do some benchmarking... the macro.underscore version is approximately 2x faster... will keep it for now. thx for the proposal!