Refactoring for better Prometheus' logic support

1) Breaking changes! Updated collection code for proper Prometheus support
- Now metrics names of all subcollectors have prefixes:
  nsq_topic_, nsq_channel_, nsq_client_
  As a result metrics registration code now passing Prometheus' client validation,
  also it fixes scrape warnings
- "type" label in exported metrics removed as obsolete
2) Add reset of gauges before scrape (like in haproxy-exporter)
3) Refactor: subcollectors logic is simplified, multiple collectors support removed
   We always use 'stats' collector so it removes unnecessary flexibility and complexity.

collector/collector.go

@@ -2,7 +2,11 @@ package collector
 
 import "github.com/prometheus/client_golang/prometheus"
 
-// Collector defines the interface for collecting all metrics for Prometheus.
-type Collector interface {
-	Collect(out chan<- prometheus.Metric) error
+// StatsCollector defines an interface for collecting specific stats
+// from a nsqd exported stats data.
+type StatsCollector interface {
+	set(s *stats)
+	collect(out chan<- prometheus.Metric)
+	describe(ch chan<- *prometheus.Desc)
+	reset()
 }