Improve shift documentation

[Shnatsel]

See also: #163

Doesn't document #280 since that's hopefully temporary

[LaurenzV]

LGTM! Though it's still an open question whether we should keep the overload (which is what the issue was mainly about I think), any opinion that? I guess people might not discover the doc comment if they just use << directly.

[Shnatsel]

I genuinely don't know.

std::simd uses << but it's not exactly a model of performance.

SSE4.2 is 15% of x86 in active use, only 8-bit AVX2 will be slow forever, so I guess it's not that bad?

I can't come up with a good name for it either. shift is close to shift_elements and is easy to confuse. The safe_arch crate uses shl_each_* but safe_arch isn't exactly widely used, and even wide built on top doesn't expose those ops at all.

[Shnatsel]

We might be able to do better for u8 than outright scalarizing with some clever emulation.

Sadly this optimization work is blocked by multiple compiler bugs:

rust-lang/rust#159831 rust-lang/rust#159464