Implement IntoValues and inner converter for Map

[tisonkun]

Hi @dtolnay!

I'm using the map.rs file in my personal project today, and I made some changes that may be useful for the upstream:

impl<K, V> From<MapImpl<K, V>> for Map<K, V> {
    fn from(map: MapImpl<K, V>) -> Self {
        Map { map }
    }
}

impl Map<String, Value> {
    /// Consumes the `Map`, returning the underlying Map implementation.
    pub fn into_inner(self) -> MapImpl {
        self.map
    }

    /// Gets an iterator over the values of the map.
    #[inline]
    pub fn into_values(self) -> IntoValues {
        IntoValues {
            iter: self.map.into_values(),
        }
    }
}

/// An owning iterator over a serde_json::Map's values.
pub struct IntoValues {
    iter: IntoValuesImpl,
}

#[cfg(not(feature = "preserve_order"))]
type IntoValuesImpl = btree_map::IntoValues<String, Value>;
#[cfg(feature = "preserve_order")]
type IntoValuesImpl = indexmap::map::IntoValues<String, Value>;

delegate_iterator!((IntoValues) => Value);

I'm willing to send a PR if this is desired to be included in the upstream.

Besides, the clippy workaround for derivable_impls seems resolved now. I remove it and no more new warnings.

[dtolnay]

into_values would be fine. The other 2 APIs in your suggestion would be misguided to add in serde_json. Either of those makes indexmap upgrades such as #1031 only doable across a major version of serde_json which is too disruptive.

[tisonkun]

@dtolnay Thanks for your information and that's a good point!

Submitted a PR at #1226.

Out of the thread (for sharing):

1. I found this patch when developing SPath (a JSONPath impl), while I later noticed that I can write a trait rather than making another Value struct.
2. Such an into_xxx function reminds me of the other PR I made Impl String::into_chars rust-lang/rust#133057. I just often find that if we can take the ownership, we can save a clone. But generally speaking, "move"s are often/always "memcpy" in Rust?