Optimize vec::retain performance
This simply moves the loops into the inner function which leads to better results.
```
old:
test vec::bench_retain_100000 ... bench: 203,828 ns/iter (+/- 2,101)
test vec::bench_retain_iter_100000 ... bench: 63,324 ns/iter (+/- 12,305)
test vec::bench_retain_whole_100000 ... bench: 42,989 ns/iter (+/- 291)
new:
test vec::bench_retain_100000 ... bench: 42,180 ns/iter (+/- 451)
test vec::bench_retain_iter_100000 ... bench: 65,167 ns/iter (+/- 11,971)
test vec::bench_retain_whole_100000 ... bench: 33,736 ns/iter (+/- 12,404)
```
67180ef056