std: use ptr.offset where possible in the vec iterator.

[huonw]

Closes#8212.

Before:

%sum.028 = phii64 [ %23, %match_else ], [ 0, %"normal return" ]
%.sroa.0.027 = phii64* [ %21, %match_else ], [ %12, %"normal return" ]
%19 = ptrtointi64*%.sroa.0.027toi64%20 = addi64%19, 8%21 = inttoptri64%20toi64*%22 = loadi64*%.sroa.0.027, align8%23 = addi64%22, %sum.028%24 = inttoptri64%20to %"enum.std::libc::types::common::c95::c_void[#1]"*
%25 = icmpeq %"enum.std::libc::types::common::c95::c_void[#1]"* %24, %16%26 = icmpeqi64%20, 0%or.cond = ori1%25, %26

test vec::bench::iterator ... bench: 87 ns/iter (+/- 7)
test vec::bench::mut_iterator ... bench: 87 ns/iter (+/- 10)

After:

%sum.028 = phii64 [ %21, %match_else ], [ 0, %"normal return" ]
%.sroa.0.027 = phii64* [ %19, %match_else ], [ %12, %"normal return" ]
%19 = getelementptri64*%.sroa.0.027, i641%20 = loadi64*%.sroa.0.027, align8%21 = addi64%20, %sum.028%22 = bitcasti64*%19to %"enum.std::libc::types::common::c95::c_void[#1]"*
%23 = icmpeq %"enum.std::libc::types::common::c95::c_void[#1]"* %22, %16%24 = icmpeqi64*%19, null%or.cond = ori1%23, %24

test vec::bench::iterator ... bench: 79 ns/iter (+/- 8)
test vec::bench::mut_iterator ... bench: 78 ns/iter (+/- 2)

(NB. I'm not 100% sure the benchmarks were set-up correctly, so take them with a grain of salt.)

[huonw]

r? @thestinger

[thestinger]

\o/

[thestinger]

@huonw: it's fine if the benchmark doesn't show a statistically significant improvement, because I didn't expect any improvement in iteration speed alone - just when combined with other things (it can now optimize iterators to a memset/memcpy with loop-idiom, vectorize them with loop-vectorize and so on).

[huonw]

@thestinger yes, it was just a warning to anyone that this shouldn't be trumpeted as a 10% speed-up without further investigation.

[thestinger]

Landing with #8257.