Possible pipeline.Discard bug

[derekperkins]

At the beginning of the Discard() function, it checks to see if the context has been canceled and errors out if it has

olric/pipeline.go

Lines 441 to 448 in 4076aef

	func (dp *DMapPipeline) Discard() error {
	select {
	case <-dp.ctx.Done():
	return ErrPipelineClosed
	default:
	}
	dp.cancel()

However the same context is canceled during Exec(), so Discard() will always return an error afterwards, which is its primary use case

olric/pipeline.go

Lines 405 to 416 in 4076aef

	// Exec executes all queued commands using one client-server roundtrip.
	func (dp *DMapPipeline) Exec(ctx context.Context) error {
	select {
	case <-dp.ctx.Done():
	return ErrPipelineClosed
	default:
	}
	dp.mtx.Lock()
	defer dp.mtx.Unlock()
	defer dp.cancel()

I think what needs to happen is to have two separate contexts / channels:

1. Used as it is today to block in FutureXXX to wait for Exec to run
2. A second that is closed inside of Discard and Close that marks all FutureXXX from that pipeline as invalid, rather than potentially returning invalid data / panicking as they would today

[derekperkins]

Fixed by #206