fix: allow any int type for indexing

[henryiii]

This expands the current panic if the index is out of bounds into panic if a negative index is given, and allows negative indexes (along with other bit lengths).

[becheran]

Bit similar to the discussion in #60

I am not sure if adding this feature would be in the scope of the goal of mimicing vec, but in a 2D data object.

The added try_into code will also add a bit of runtime overhead. Although, to be fair, it will be fairly low. Still. I am not sure if using an unsigned index is really that much of requirement. It would also be fairly simply to convert from signed to unsigned before indexing on the calling side and handle issues such as overflows outside of grid.

[henryiii]

I was expecting that try_into should be implemented as nothing and optimize away for matching sizes. The extra cost should only occur if it's converting. (Though I agree if that's not the case!)

[henryiii]

It looks like the ASM is different, I tried these two:

use core::ops::Index;

struct A {
    a: Vec<u8>,
}

impl<T: TryInto<usize>> Index<(T, T)> for A {
    type Output = u8;

    #[inline]
    fn index(&self, (row, col): (T, T)) -> &u8 {
        &self.a[row.try_into().ok().unwrap()*col.try_into().ok().unwrap()]
    }
}

/*
struct A {
    a: Vec<u8>,
}

impl Index<(usize, usize)> for A {
    type Output = u8;

    #[inline]
    fn index(&self, (row, col): (usize, usize)) -> &u8 {
        &self.a[row*col]
    }
}
*/

fn main() {
    let v = A{a: vec![0,1,2,3,4,5,6,7]};
    println!("{}", v[(3,2)]);
}