Make the event processing chain configurable

[dxsup]

Currently, we build the processing chain by hard coding, which makes changes inconvenient. Every time we add a new component to the chain, we have to modify the source codes and make the building procedure more verbose and difficult to read. What looks like this:

// buildPipeline builds a event processing pipeline based on hard-code.
func (a *Application) buildPipeline() error {
	// TODO: Build pipeline via configuration to implement dependency injection
	// Initialize exporters
	otelExporterFactory := a.componentsFactory.Exporters[otelexporter.Otel]
	otelExporter := otelExporterFactory.NewFunc(otelExporterFactory.Config, a.telemetry.Telemetry)
	// Initialize all processors
	// 1. Kindling Metric Format Processor
	formatProcessorFactory := a.componentsFactory.Processors[kindlingformatprocessor.ProcessorName]
	formatProcessor := formatProcessorFactory.NewFunc(formatProcessorFactory.Config, a.telemetry.Telemetry, otelExporter)
	// 2. Kubernetes metadata processor
	k8sProcessorFactory := a.componentsFactory.Processors[k8sprocessor.K8sMetadata]
	k8sMetadataProcessor := k8sProcessorFactory.NewFunc(k8sProcessorFactory.Config, a.telemetry.Telemetry, formatProcessor)
        // other initialization
        ...
	a.analyzerManager = analyzerManager
	udsReceiverFactory := a.componentsFactory.Receivers[udsreceiver.Uds]
	udsReceiver := udsReceiverFactory.NewFunc(udsReceiverFactory.Config, a.telemetry.Telemetry, analyzerManager)
	a.receiver = udsReceiver
	return nil
}

Although our processing chain is inspired by the pipeline of Opentelemetry-Collector, its building procedure can't be applied to ours directly. Because the pipeline is one direction for every processor while the processing chain is more like a directed acyclic graph in which every node has multiple fan-out directions.

I will work on this issue.