fix: use static tokio runtime to prevent shutdown during sync

[raine]

Fixes #4040.

Problem

The rt() function creates a new Tokio runtime on each call, returns its Handle, and immediately drops the runtime. This causes panics during sync operations:

thread panicked at: A Tokio 1.x context was found, but it is being shutdown.

Solution

Use OnceLock to create a single static runtime that lives for the program's lifetime. Also use enable_all() to enable both time and IO features needed for sync operations.

Changes

// Before: runtime dropped immediately after getting handle
fn rt() -> tokio::runtime::Handle {
    tokio::runtime::Builder::new_current_thread()
        .build()
        .unwrap()
        .handle()
        .clone()
}

// After: static runtime lives for program lifetime
use std::sync::OnceLock;

static RUNTIME: OnceLock<tokio::runtime::Runtime> = OnceLock::new();

fn rt() -> &'static tokio::runtime::Runtime {
    RUNTIME.get_or_init(|| {
        tokio::runtime::Builder::new_current_thread()
            .enable_all()
            .build()
            .unwrap()
    })
}

Note: PR #4042 adds .enable_time() which addresses the "timers are disabled" error but doesn't fix the underlying runtime lifetime issue.

[djmitche]

Oops, I just saw this after merging #4042. This looks like a better fix! Why close it?

[djmitche]

I will rebase and adopt this! (and figure out how to test this more effectively)

[djmitche]

I confirmed that with this change applied my local GCP sync works fine.

[djmitche]

Thanks!