Inline Finallyalizer::drop, allowing LLVM to optimize finally.

[eddyb]

• fixes the vec::from_elem regression caused by std: Make vec::from_elem failure-safe #8780
• added 5 benchmarks for allocating a 1KB ~[u8] and zeroing it
• closes vec::from_elem with primitives should be as fast as calloc #7136

[eddyb]

cc @cmr @huonw

[eddyb]

r?

/* Reference result (TESTNAME=zero_1kb):
test vec::bench::zero_1kb_fixed_repeat                          ... bench:       240 ns/iter (+/- 16)
test vec::bench::zero_1kb_from_elem                             ... bench:       237 ns/iter (+/- 32)
test vec::bench::zero_1kb_loop_set                              ... bench:      1581 ns/iter (+/- 37)
test vec::bench::zero_1kb_mut_iter                              ... bench:       239 ns/iter (+/- 38)
test vec::bench::zero_1kb_set_memory                            ... bench:       236 ns/iter (+/- 9)
*/