Add multiple types of application metrics

[ccicconetti]

Functions can cast events containing run-time metrics to the channel metric.

If there's a metrics-collector resource provider in the cluster, it will collect the samples and make them available to the ε-ORC, which will save them to a Redis if configured to do so.

This extends and substitutes the current implementation, which only supports workflow/function latencies.

Types of metrics to be implemented:

• [scalar] gauges: last value is overwritten (example: current inference speed)
• [scalar] counters: value is added (example: number of objects detected)
• [list] durations: a sample is added with the difference between ts_end and ts_begin (example: transaction latency); a scalar value with the EWMA is added automatically
• [list] single occurrences: a new sample is added each time cast() is invoked (example: an anomaly was detected)

All the values are f64 and are assigned a timestamp and an identifier local to the resource stream.