Introduce a mechanism for contextualizing panics

[cole-miller]

Introduces a way for callsites throughout the Zed client to pass context to our custom panic hook (defined in crates/zed/src/reliability.rs) and ultimately to #panics in the Zed Slack. When we see a particular panic backtrace repeatedly and the cause isn't immediately obvious, this allows us to update the code to package some additional contextual information when that panic occurs in the future. It combines well with catch_unwind for cases where we want to stop the panic from crashing Zed but still get a telemetry report about it, since the panic hook is called even for panics that are caught. Usage:

// Don't catch the panic, just propagate it.
let result = {
    let _guard = reliability::hook_fn({
        let gizmo = gizmo.clone();
        move |_panic_info| format!("gizmo state: {gizmo:?}")
    });
    gizmo.panicky_operation()
}

// Catch the panic but still send a report.
let result: std::thread::Result<T> = reliability::hook_fn(|_| format!("context!")).catch_unwind(|| gizmo.panicky_operation());

// Catch the panic from polling a future.
let result: std::thread::Result<T> = reliability::hook_fn(|_| format!("context!")).catch_unwind_future(gizmo.panicky_async_operation()).await;

Release Notes:

• N/A