Bad custom test names should not fail silently

I will be borrowing the example from #19, since it has some relation to how I found this error on my own code.

So, we know from that issue that the minus will be abstracted from the automatic name creation process, BUT, we have a mechanism for custom names, and that could help the developer to understand that the problem with his code is that he is lacking a valid test name:

```
#[test_case(15, 15; "Abs_15_15")]
#[test_case(-15, 15; "Abs_-15_-15")]
fn test_crazy_stuff(value: i32, expected: i32) {
assert_eq!(value.abs(), expected)
}
```

And this will raise a familiar error, but the interesting part is that it tried to use my custom name, since it begins with `abs`, and not `test_crazy_stuff`:

```
error[E0428]: the name `abs_15_15` is defined multiple times
--> src/lib.rs:162:5
|
162 | #[test_case(15, 15; "Abs_15_15")]
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
| |
| `abs_15_15` redefined here
| previous definition of the value `abs_15_15` here
```

This leads me to believe that we have a process that tries to make the name valid, since it erased the hyphen from my custom test, applied, and then we had a name collision.

I haven't tried to analyze the code to understand how the hyphen is erased in both cases, but, i believe that if a developer instantiates a custom name, it shouldn't be safeguarded, if the name is invalid it should break catastrophically warning the developer of what the problem is.