Manually implement Clone

[AntonSol919]

I noticed clone was producing inefficient assembly.
Here is a PR that manually implements Clone.

The first commit has benchmarks comparing three methods.

Here are my results:

Cloning inline strings:

running 3 tests
test bench::bench_derive_clone ... bench:     460,725 ns/iter (+/- 9,990)
test bench::bench_match_clone  ... bench:     177,981 ns/iter (+/- 13,023)
test bench::bench_new_clone    ... bench:     178,105 ns/iter (+/- 4,005)

Cloning heap strings:

running 3 tests
test bench::bench_derive_clone ... bench:   3,197,654 ns/iter (+/- 98,527)
test bench::bench_match_clone  ... bench:   3,193,360 ns/iter (+/- 74,968)
test bench::bench_new_clone    ... bench:   3,192,459 ns/iter (+/- 76,398)

[Veykril]

match vs unsafe does not seem to really make a difference here does it? In that case I'd probably prefer a safe implementation via matching. Otherwise, at least instead of transmute_copy I would want to use ptr::read here, as we aren't transmuting.

[AntonSol919]

Match also uses unsafe. Its the only method I know to prevent the compiler from emitting additional branches.

I updated the PR to use ptr read.

Note that f8ed961 caused a regression due to Repr ordering (see the first commit for my results).

[Veykril]

Needs a reformatting

[Veykril]

Thanks!