Fix std::fmt::Display for abort expressions

[DylanRJohnston]

Fixes #829. I couldn't see any tests for this code, but a very simply one exists as I wrote in the issue. Let me know if you want that included and where it should live.

    #[test]
    fn abort_display_isomorphism() {
        let expression = r#"abort "error message""#;

        assert_eq!(
            format!("{expression}\n"),
            vrl::parser::parse(expression).unwrap().to_string()
        )
    }

[DylanRJohnston]

Just noticed a few more instances of failing to serialize out re-parsable expressions, namely infailable function calls and queries.

[DylanRJohnston]

Without this variable queries end up mushed together e.g. foo.bar becomes foobar however that then causes .foo.bar to become ..foo.bar without the change to QueryTarget below.

[DylanRJohnston]

Should probably just remove this write! all together? If I understand correctly, and based on my testing QueryTarget always appears with a . after it even if OwnedValuePath is empty. So this is safe.

[pront]

Hi @DylanRJohnston, your project sounds quite interesting but this PR breaks a test. I don't have time to do a thorough review but for faster dev iterations you can locally use ./scripts/checks.sh or just ./scripts/checks.sh vrl_tests tests.

[pront]

Hi @DylanRJohnston, since this is deep change, it is also worth pointing Vector's VRL dependency to this sha and see if those CI checks are passing.

[DylanRJohnston]

Hey @pront apologies I was waiting until the project was complete before coming back to fixing this pull request as I keep uncovering new problems as the project progresses and I didn't want to waste everyone's time with lots of pull requests. Since these changes are only needed at transpile time I've been able to get away with just using a fork for now although I am very interested in getting something like these changes merged into upstream.

I also made a fairly controversial decision to overload the fmt::Formatter width property to carry the indentation level for pretty printing of nested scopes and I wasn't sure you guys would approve of abusing it like that although I can't see any other way other than writing a whole separate serialization strategy that could carry information like that instead of the current reliance on recursive invocation of fmt::Display.

I also want to write some more comprehensive tests that ensure the parse <-> serialize isomorphism beyond just the edge cases I've discovered assuming this is a property that the VRL project wants to maintain?