Remove wrong &str + String and &[T] + Vec<T> implementations

[japaric]

TL;DR I wrongly implemented these two ops, namely "prefix" + "suffix".to_string() gives back "suffixprefix". Let's remove them.

The correct implementation of these operations (lhs.clone().push_str(rhs.as_slice())) is really wasteful, because the lhs has to be cloned and the rhs gets moved/consumed just to be dropped (no buffer reuse). For this reason, I'd prefer to drop the implementation instead of fixing it. This leaves us with the fact that you'll be able to do String + &str but not &str + String, which may be unexpected.

r? @aturon
Closes #19952

[aturon]

I'm not sure I agree that the potential inefficiency of a particular operation is a good reason to lose the symmetry of addition. People doing string manipulation where this kind of thing matters should probably be working more explicitly in the first place; these operators are mainly for convenience.

Also, it's feasible to reuse the buffer if you wanted -- basically a bit like the insert operation on Vec, but inserting a whole slice on the left. You could special-case this behavior depending on whether the existing capacity was sufficient, and otherwise allocate a fresh buffer and copy each bit in.

[alexcrichton]

Ah yes I should have commented to that effect as well. I'm ok with removing these for now, but I'd like to add them back at some point. You can always reserve extra space, slide everything to the right in the owned container, and then clone the slice into the empty space for an "optimal" solution I believe.

[renato-zannon]

Having a String#prepend method that works as @alexcrichton described sounds useful regardless of the symmetry of addition.