bug: BatchQuery executor read with expired epoch under parallel e2e

[BugenZhao]

Recently we've enabled the watermark checking, and some bugs of reading with expired epochs are revealed. The batch query executor may read with an expired epoch under parallel e2e, which might be caused by the epoch of creating mview failing to be pinned.

https://buildkite.com/risingwavelabs/pull-request/builds/9338#01837ead-70bb-4053-8d84-d9ef51e23c5f

The unpin here might be problematic. With concurrent DDL, a calling of unpin may also unpin the snapshot for other creating mviews.

risingwave/src/meta/src/barrier/schedule.rs

Lines 185 to 199 in e081731

	// TODO: refactor this
	if is_create_mv {
	// The snapshot ingestion may last for several epochs, we should pin the epoch here.
	// TODO: this should be done in `post_collect`
	let _snapshot = self.hummock_manager.pin_snapshot(META_NODE_ID).await?;
	// Wait for this command to be finished.
	let res = finish_rx.await;
	self.hummock_manager.unpin_snapshot(META_NODE_ID).await?;
	res.map_err(|e| anyhow!("failed to finish command: {}", e))?;
	} else {
	// Wait for this command to be finished.
	finish_rx
	.await
	.map_err(|e| anyhow!("failed to finish command: {}", e))?;
	}