flush frames and confirm snapshot completion before calling SQLite checkpoint

[Horusiath]

This solves the issue when SQLite files (DB + WAL) are being modified before we confirm that they were backed up.

PS: don't merge yet, there's still an ongoing issue with tokio task never being scheduled causing a deadlock on subsequent checkpoints:

sqld/bottomless/src/replicator.rs

Lines 751 to 789 in dcb5840

	let handle = tokio::spawn(async move {
	let start = Instant::now();
	let body = match Self::maybe_compress_main_db_file(db_file, compression).await {
	Ok(file) => file,
	Err(e) => {
	tracing::error!(
	"Failed to compress db file (generation {}): {}",
	generation,
	e
	);
	let _ = snapshot_notifier.send(Err(e));
	return;
	}
	};
	let mut result = snapshot_req.body(body).send().await;
	if let Err(e) = result {
	tracing::error!(
	"Failed to upload snapshot for generation {}: {:?}",
	generation,
	e
	);
	let _ = snapshot_notifier.send(Err(e.into()));
	return;
	}
	result = change_counter_req.send().await;
	if let Err(e) = result {
	tracing::error!(
	"Failed to upload change counter for generation {}: {:?}",
	generation,
	e
	);
	let _ = snapshot_notifier.send(Err(e.into()));
	return;
	}
	let _ = snapshot_notifier.send(Ok(Some(generation)));
	let elapsed = Instant::now() - start;
	tracing::debug!("Snapshot upload finished (took {:?})", elapsed);
	let _ = tokio::fs::remove_file(format!("db.{}", compression)).await;
	});