Support SystemInput tuples up to 8 elements

[ItsDoot]

Objective

• Writing an API, and I want to allow users to pass in extra data alongside the API provided input, and tuples are the most natural extension in this case.
• Bring SystemInput up to par with SystemParam for tuple support.

Solution

• Added impls for tuples up to 8 elements. If you need a 9-arity tuple or more, write your own SystemInput type (it's incredibly simple to do).

Testing

• Added a test demonstrating this.

---

Showcase

Tuples of arbitrarySystemInputs are now supported:

fn by_value((In(a), In(b)): (In<usize>, In<usize>)) -> usize {
    a + b
}
fn by_mut((InMut(a), In(b)): (InMut<usize>, In<usize>)) {
    *a += b;
}

let mut world = World::new();
let mut by_value = IntoSystem::into_system(by_value);
let mut by_mut = IntoSystem::into_system(by_mut);

by_value.initialize(&mut world);
by_mut.initialize(&mut world);

assert_eq!(by_value.run((12, 24), &mut world), 36);

let mut a = 10;
let b = 5;
by_mut.run((&mut a, b), &mut world);
assert_eq!(*a, 15);

[alice-i-cecile]

@spectria-limina you were poking at this; can I get your review here?

[spectria-limina]

LGTM

[alice-i-cecile]

This should also be documented on In.

[ItsDoot]

In used to be the only way to pass input into systems, but given it's now just one of the multiple SystemInput types, I don't think In needs to be treated any differently from the others. I have added an extra note on all of the input types to check the trait for more information, however. I also added an example to the docs of using tuples.

[alice-i-cecile]

Surprisingly simple and well-tested. I think we can probably do a bit better with docs.

[bushrat011899]

Minor gripe around max tuple size documentation that isn't even specific to this PR. Otherwise, this just looks like a nice change!

[bushrat011899]

Not a problem with this PR specifically, but I feel like Bevy should probably just have a constant somewhere for LONGEST_REASONABLE_TUPLE, instead of each site picking a somewhat random value (I think I've seen 12, 15, and 16, now also 8).

[ItsDoot]

That's fair, I just think its way less reasonable to have large-arity input tuples when even using system input at all is somewhat uncommon. So I'd prefer if we could (potentially) save some compilation time by not implementing for arities that are highly unlikely to be used in practice.