Use serde_core where possible

[AdamGS]

Using the newly released serde_core seems to have significant benefit for compile times, pbjson is the main user left in Datafusion.

This is a breaking change, as it generates different code (serde_core:: instead of serde::), and also significantly moves up the required serde/serde_core versions.

There is one #[derive(Deserialize)] which I couldn't figure out yet, but I figured its worth opening this PR for early feedback.

[AdamGS]

Seems like reproducing the auto-generated code requires something like serde-content, would that be an acceptable addition here?

[crepererum]

Seems like reproducing the auto-generated code requires something like serde-content, would that be an acceptable addition here?

Can you elaborate why this is required? I'm failing to see the reason for that 😅

[crepererum]

OK, I did look into this more in depth: the issue is pbjson::private::Content and the borrowing deserialization that comes with it. If you cargo-expand the derived implementation, you end up with a lot of private serde APIs used. So we would either implement the glue code ourselves (a few hundred lines) or use -- as @AdamGS suggested -- serde-content. It's a bit of a shame that the serde_core crates are basically unusable without helpers for more complex settings.