Make AudioFrame trivially copyable, and remove extraneous warning delints

[Ivorforce]

What problem(s) does this PR solve?

AudioFrame is currently not trivially copyable.
While perhaps optimizable in later compiler stages, it is unnecessary to rely on this. It should be cleaned up.

Additional information

I also made Vector2 conversions explicit. Besides being a core goal, Vector2 uses real_t while AudioFrame uses float so there's real risk of data loss.

[clayjohn]

Seems that this code was intentional. This PR basically brings us back to the state before #87006

[Ivorforce]

Seems that this code was intentional. This PR basically brings us back to the state before #87006

It doesn't, it was l and r before; it's left and right now (still resolving the problem).

The original PR is a (small) performance hazard in hindsight (making this type not trivially copyable and not constexpr compatible) and should not have been merged in the form that it was.

[hpvb]

The exact right amount of cursed.

[hpvb]

I retract the below statement. After talking it through with @Ivorforce it seems obvious that the standard already guarantees for the cases we care about. Thanks for humoring me!

Details

Looking at this a bit more (because I was reviewing #120845), I think maybe we should add one more static assert to this (And all of the changes in #120845) It is perhaps super pedantic, but what else am I here for:

static_assert(alignof(AudioFrame) >= alignof(float));

Like, in practice I don't think there's really any case where this can't hold, but if we're doing the static asserts anyway...

[Ivorforce]

We resolved the above comment in chat. Practically, the guarantee follows from the "first member is at offset 0" guarantee + the "every member is placed at its correct alignment" guarantee. This can only hold if the owning struct's alignment is at least that of each of its fields.

[Repiteo]

Thanks!