Allow external tracing procedures

[ma2bd]

Summary

Following suggestions from @jordens in #73, we make a number of APIs and types pub.

Test Plan

CI

[jordens]

Thanks.
While this would work for me, I rejected the same approach in my PR because it (1) leaves the door open for an inconsistent registry (missing variants) and (2) misses the opportunity to detect an infinite loop due to logic errors that the other approach could detect. The latter is merely a side effect as there is no known case of such an infinite loop AFAICT. The first one is the original argument. See below for a unittest showing this problem.

[jordens]

With this you can do this (the reduced test from #73):

#[test]
fn test_pending_enum() {
    #[derive(Deserialize)]
    enum E {
        A,
        B,
    }


    let mut tracer = Tracer::new(TracerConfig::default());
    let samples = Samples::new();
    let mut format = Format::unknown();
    let deserializer = Deserializer::new(&mut tracer, &samples, &mut format);
    E::deserialize(deserializer).unwrap();
    format.reduce();
    let Format::TypeName(name) = &format else {
        panic!()
    };
    assert!(tracer.check_incomplete_enum(name).is_some());
    assert_eq!(
        tracer.registry(),
        Err(Error::MissingVariants(vec!["E".to_owned()]))
    );
}

which results in:

---- test_pending_enum stdout ----
thread 'test_pending_enum' panicked at serde-reflection/tests/serde.rs:527:5:
assertion `left == right` failed
  left: Ok({"E": Enum({4294967294: Named { name: "A", value: Unit }})})
 right: Err(MissingVariants(["E"]))

[jordens]

ma2bd/serde-reflection@external_tracing...quartiq:serde-reflection:external_tracing If you want to play with it.

[ma2bd]

Let's start with this PR and see if we need anything more later.