Performance: change user interface to reduce allocs

[ShimmerGlass]

The previous implementation, while user friendly, required a lot of allocations on the hot path. This changes it for an iterator based implementation that, along with a few smaller optimization, reduces the allocs by 82% and improves latency by 25%.

Benchmarks comparison:

benchmark           old ns/op     new ns/op     delta
BenchmarkSPOE-6     32842         24653         -24.93%

benchmark           old allocs     new allocs     delta
BenchmarkSPOE-6     23             6              -73.91%

benchmark           old bytes     new bytes     delta
BenchmarkSPOE-6     1079          193           -82.11%

This branch is based on timeouts for convenience. I'll rebase it afterwards.

[ShimmerGlass]

@pierresouchay, ya there is no version currently for this lib. I'm planning on releasing v1.0.0 after these two are merged

[pierresouchay]

@ShimmerGlass There are confllicts

[ShimmerGlass]

@pierresouchay rebased