test: Add BenchmarkEncodeDecode

[peterhellberg]

Initial benchmarking of sqids.Encode and sqids.Decode for the input []uint64{1, 2, 3, 4, 5} on my particular machine gives me these numbers:

$ time go test -bench=. -run=BenchmarkEncodeDecode -benchmem -cpuprofile profile.out
goos: linux
goarch: amd64
pkg: github.com/sqids/sqids-go
cpu: Intel(R) Core(TM) i7-10700T CPU @ 2.00GHz
BenchmarkEncodeDecode-16           12253         97055 ns/op       32470 B/op        124 allocs/op
PASS
ok      github.com/sqids/sqids-go   2.214s

real    0m2.370s
user    0m2.299s
sys 0m0.175s

pprof shows us that a lot of unnecessary time is spent in dealing with a regular expression:

[!Note] (Re-)Compilation (and usage) of regular expressions are pretty slow in Go, so should be avoided if possible.

[peterhellberg]

With the change in 4e29f28 the benchmark output on my machine change to:

$ time go test -bench=. -run=BenchmarkEncodeDecode -benchmem -cpuprofile profile.out
goos: linux
goarch: amd64
pkg: github.com/sqids/sqids-go
cpu: Intel(R) Core(TM) i7-10700T CPU @ 2.00GHz
BenchmarkEncodeDecode-16           27223         42703 ns/op       31474 B/op        115 allocs/op
PASS
ok      github.com/sqids/sqids-go   1.810s

real    0m2.017s
user    0m1.986s
sys 0m0.199s

So a difference of 97055 ns/op to 42703 ns/op

And now pprof shows us that more time is spent in shuffle than in isBlockedID

[4kimov]

Whoa. Nice 💪

[peterhellberg]

@4kimov Is it cool if I merge PRs or do you want the final say in what is going into the library?

[4kimov]

@peterhellberg No, I'm cool with you merging, I trust your best judgement.