Optimize LocalVector::push_back for non-trivial objects

[mihe]

I happened to stumble upon this discrepancy between how trivially constructible objects and non-trivially constructible objects were being constructed in LocalVector::push_back, where trivially constructible objects seemed to be using std::move (thus avoiding the redundant copy) and non-trivially constructible objects were not.

It seemed to be something that was missed as part of #100477, as mentioned by @Ivorforce here.

Some quick for-loop profiling using this code with MSVC...

constexpr int N = 100000;
constexpr int M = 8;
LocalVector<Variant> vector;
vector.reserve(N * M);

Variant variants[M] = {
	true,
	42,
	42.0f,
	String("Lorem ipsum dolor sit amet, consectetur adipiscing elit."),
	Vector3(),
	Transform2D(),
	Transform3D(),
	Projection()
};

auto t0 = std::chrono::high_resolution_clock::now();
for (int i = 0; i < N; ++i) {
	for (int j = 0; j < M; ++j) {
		vector.push_back(variants[j]);
	}
}
auto t1 = std::chrono::high_resolution_clock::now();
std::cout << std::chrono::duration_cast<std::chrono::nanoseconds>(t1 - t0).count() << "ms\n";

... yielded 15.48 ms without the change and 12.89 ms with the change, so roughly a 16% reduction for Variant specifically, but obviously depends greatly on what the object does in its copy constructor.

(This also seems to be the only thing standing in the way of LocalVector being able to contain move-only types.)

[Ivorforce]

Looks correct. I merely forgot to add it. Thanks for stepping in!

[mihe]

I added the 4.4 cherrypick label to it, since I figured it's a small change and fairly low risk, but I'm honestly not sure what the general stance is on cherrypicking performance improvements like this.

[lawnjelly]

Just checked and I had already added this in 3.x with #100995 , and it was backporting @Ivorforce PRs, so maybe it was there somewhere originally for 4.x (but got missed somehow).

UPDATE: Ah no, Lukas noticed it as an addition, but it never got added to 4.x.

[clayjohn]

I added the 4.4 cherrypick label to it, since I figured it's a small change and fairly low risk, but I'm honestly not sure what the general stance is on cherrypicking performance improvements like this.

Usually we don't, but since this change is really to complete a PR from 4.4, I think it makes sense

[akien-mga]

Thanks!