Improve errors upon deserializing NonZero*

[fulara]

Deserializing errors on NonZero* could definitely be improved, not sure if this is possible, but somehow the normal numerical types have better information attached to them - is this fault of the wrappers - probably not.

silly example code

#[derive(Deserialize)]
struct Example {
    a: i32,
    b: i32,
    c: i32,
    d: i32,
    e: i32,
    f: i32,
    g: i32,
    h: i32,
    i: i32,
    j: i32,
    k: i32,
    l: i32,
    m: i32,
    n: i32,
    non_zero_usize: NonZeroUsize,
    usize: usize,
}

#[derive(Deserialize)]
struct Parent {
    example: Example,
}

#[test]
fn test() {
    let x = r#"
example:
  a: 0
  b: 0
  c: 0
  d: 0
  e: 0
  f: 0
  g: 0
  h: 0
  i: 0
  j: 0
  k: 0
  l: 0
  m: 0
  n: 0
  non_zero_usize: 1
  usize: 1
"#;
    let parsed: Parent = serde_yaml::from_str(x).unwrap();
}

When setting the non_zero_usize to '0' - misleading error.

expected a non-zero value", Some(Pos { marker: Marker { index: 13, line: 3, col: 3 }, path: "example" })

When setting the usize to '-1' - Nice error!

"invalid value: integer `-1`, expected usize", Some(Pos { marker: Marker { index: 137, line: 18, col: 9 }, path: "example.usize" })

As you can passing zero to non_zero_usize causes serde not to deserialize as expected but errors are hard to diagnose.
Following area could be improved:

1. Path of the last node is swallowed, ( it says just example - unlike for usize where it says example.usize )
2. The line is pointing at god knows what confusing readers.

When a person that has no access to the codebase encounters those errors it's almost impossible to guess what's the cause of the problem

[dtolnay]

I published a fix in serde 1.0.134.