Matrix Display output alignment is not Unicode-aware

Hi,

Thanks for all the wonderful work on this library. I'm coming back after a Rust hiatus and I can't believe the const generics work as well as they do.

I'm using a custom scalar type that uses a combining overline character instead of a minus sign, to match the style of this reference work:


Somewhat ironically, a typesetting convention designed to make alignment easier in print breaks alignment for `Matrix` output using this custom type:

```rs
┌ ┐
│ 0 ̅1 0 0 │
│ 0 ̅1 0 0 │
│ ⅔ 0 1 0 │
│ 0 0 0 1 │
└ ┘
```

The `Display` implementation [here](https://docs.rs/nalgebra/latest/src/nalgebra/base/matrix.rs.html#1897) uses `.char().count()` instead of [grapheme](https://docs.rs/unicode-segmentation/latest/unicode_segmentation/struct.Graphemes.html) counting, which is what causes the misalignment seen here. I imagine this would also affect less frivolous custom `Display` impls, like Chinese characters.

`unicode-segmentation` is a large dependency to make a tiny edge case look nicer, so I'm not exactly saying this *should* be implemented differently, but I'd be interested if anyone else has had this issue or if there's a solution that doesn't require increasing the code size too massively.