colcontainer: harden ctx capture in diskQueueWriter

In ffb4666 we added more memory
accounting to the disk queue. This required capturing the context inside
the `diskQueueWriter` in order to preserve the `io.Writer` signature. We
just saw a test failure where this capture was problematic - it's
possible that when we close the disk queue as a whole (which happens
_after_ closing all operators) that the captured context contains the
tracing span that has been finished (since the relevant operator has
already been closed). To go around this issue we now update the captured
context before finishing each file.

An alternative solution could've been to use `context.Background` since
we only need the context object for memory accounting purposes (in
a single `Resize` call) so losing some contextual information there
doesn't seem like a big deal.

Release note: None

Author: yuzefovich

pkg/sql/colcontainer/diskqueue.go

@@ -597,6 +597,12 @@ func (d *diskQueue) writeFooterAndFlush(ctx context.Context) (err error) {
 			d.serializer = nil
 		}
 	}()
+	if d.writer != nil {
+		// The context that we captured when we created the diskQueueWriter
+		// might have a tracing span that has already been finished. To go
+		// around this, we capture the fresh context.
+		d.writer.ctx = ctx
+	}
 	if err := d.serializer.Finish(); err != nil {
 		return err
 	}