Improve error reporting and stack tracing

[AdamGS]

Current errors tend to be very noisy, with long chains of Arrow Error:, mix of internally and externally provided sources and messy stack traces. I ran into this blogpost by the greptime people, which outlines a pretty elegant solution with very nice outcomes, both performance wise and UX wise.

I think we should consider adopting a similar approach, either by using whatever they end up extracting into a separate crate or building it on our own.

the main goals as I see them are:

1. Keep the general ergonomics of thiserror - nice From implementations for external types, ability to specify internal Vortex errors.
2. Better and more relevant stack traces - should inform user about concrete problems and issues, without having 50 lines of tokio/datafusion/random functions
3. Keep the error type(s?) useful for Vortex users
4. Obviously good performance/binary size etc.