raw_detect perf tweaks

[jqnatividad]

• applied clippy::needless_for_each lint (see https://rust-lang.github.io/rust-clippy/master/index.html#needless_for_each)
• declare scores vec with_capacity to minimize allocs

Before:

running 4 tests
test bench_alphabet_cyrillic_calculate_scores ... bench:       1,923 ns/iter (+/- 60)
test bench_alphabet_latin_calculate_scores    ... bench:       2,706 ns/iter (+/- 151)
test bench_detect                             ... bench:   4,219,612 ns/iter (+/- 129,220)
test bench_detect_script                      ... bench:      73,365 ns/iter (+/- 3,556)

After tweaks:

running 4 tests
test bench_alphabet_cyrillic_calculate_scores ... bench:       1,917 ns/iter (+/- 68)
test bench_alphabet_latin_calculate_scores    ... bench:       2,748 ns/iter (+/- 98)
test bench_detect                             ... bench:   4,188,685 ns/iter (+/- 157,631)
test bench_detect_script                      ... bench:      69,409 ns/iter (+/- 4,946)

[Ap3lsin4k]

I confirm that setting capacity is more performant.

[greyblake]

Thank you! Sorry for the delay.