Consider removing O type parameter of serialize_with_flavor

[dtolnay]

postcard::serialize_with_flavor is written in an odd way that makes the caller restate the Flavor's Output type explicitly when using turbofish.

postcard/src/ser/mod.rs

Lines 367 to 370 in a0ef8a2

	pub fn serialize_with_flavor<T, S, O>(value: &T, storage: S) -> Result<O>
	where
	T: Serialize + ?Sized,
	S: Flavor<Output = O>,

- pub fn serialize_with_flavor<T, S, O>(value: &T, storage: S) -> Result<O>
+ pub fn serialize_with_flavor<T, S>(value: &T, storage: S) -> Result<S::Output>
  where
      T: Serialize + ?Sized,
-     S: Flavor<Output = O>,
+     S: Flavor,

[dtolnay]

I can't tell whether this was a stylistic choice or oversight. My preference is for serialize_with_flavor::<T, Cobs<Slice<'a>>>(...) over serialize_with_flavor::<T, Cobs<Slice<'a>>, &'a mut [u8]>(...), but I can see there may be valid preferences in the other direction.

[jamesmunns]

As far as I can tell - this was likely an oversight, probably an artifact implementing flavors and never stepping back and realizing that those (always?) are the same type.

I do think this would be a breaking change as you noted in #105, but I'm tagging it with the 2.0 label so I don't forget to clean this up on the next major release.