auto merge of #15638 : blake2-ppc/rust/ptr-arithmetic-chars, r=huonw

Reimplement the string slice's `Iterator<char>` by wrapping the already efficient
slice iterator.

The iterator uses our guarantee that the string contains valid UTF-8, but its only unsafe
code is transmuting the decoded `u32` into `char`.

Benchmarks suggest that the runtime of `Chars` benchmarks are reduced by up to 30%,
runtime of `Chars` reversed reduced by up to 60%.

```
BEFORE
test str::bench::char_indicesator                          ... bench:       124 ns/iter (+/- 1)
test str::bench::char_indicesator_rev                      ... bench:       188 ns/iter (+/- 9)
test str::bench::char_iterator                             ... bench:       122 ns/iter (+/- 2)
test str::bench::char_iterator_ascii                       ... bench:       302 ns/iter (+/- 41)
test str::bench::char_iterator_for                         ... bench:       123 ns/iter (+/- 4)
test str::bench::char_iterator_rev                         ... bench:       189 ns/iter (+/- 14)
test str::bench::char_iterator_rev_for                     ... bench:       177 ns/iter (+/- 4)

AFTER
test str::bench::char_indicesator                          ... bench:        85 ns/iter (+/- 3)
test str::bench::char_indicesator_rev                      ... bench:        82 ns/iter (+/- 2)
test str::bench::char_iterator                             ... bench:       100 ns/iter (+/- 3)
test str::bench::char_iterator_ascii                       ... bench:       317 ns/iter (+/- 3)
test str::bench::char_iterator_for                         ... bench:        86 ns/iter (+/- 2)
test str::bench::char_iterator_rev                         ... bench:        80 ns/iter (+/- 6)
test str::bench::char_iterator_rev_for                     ... bench:        68 ns/iter (+/- 0)
```

Note: Branch name is no longer indicative of the implementation.

diff --cc src/libcollections/str.rs
Simple merge