Support collecting unsampled 100% metrics derived from traces regardless of the trace sampling rate

[ntyrewalla]

Several customers are looking for unsampled metrics that are derived from spans. These metrics include error, faults, throttle, latency, and total request counts. Also, we need to look into the aggregation level for these metrics. Based on the configuration, it can be aggregated and batched at SDK/collector at 1 min/5 mins (or any other interval) to manage the data sent across the wire.