safekeeper: eagerly return AppendResponse with flush_lsn on segment fsync

[erikgrinaker]

When pipelining AppendRequests, the WAL acceptor only returns an AppendResponse with the new flush_lsn when it itself explicitly requests a flush via FlushWAL:

neon/safekeeper/src/receive_wal.rs

Lines 573 to 580 in 0d5a512

	// While receiving AppendRequests, flush the WAL periodically and respond with an
	// AppendResponse to let walproposer know we're still alive.
	_ = flush_ticker.tick(), if dirty => {
	dirty = false;
	self.tli
	.process_msg(&ProposerAcceptorMessage::FlushWAL)
	.await?
	}

However, the WAL is also implicitly flushed on segment bounds:

neon/safekeeper/src/wal_storage.rs

Lines 308 to 310 in f18aa04

	if xlogoff + buf.len() == self.wal_seg_size {
	// If we reached the end of a WAL segment, flush and close it.
	self.fdatasync_file(&file).await?;

We should eagerly return an AppendResponse in this case, to allow the compute to commit up to that point.