Fix goroutine leak on config reload

[camathieu]

Promxy is leaking goroutines on config reloads leading to OOM kills in the long run.

From : https://instance-that-reload-its-config-way-too-often/debug/pprof/goroutine?debug=1

goroutine profile: total 63342
21111 @ 0x4374c0 0x447763 0x9bb257 0x467571
#	0x9bb256	github.com/prometheus/prometheus/discovery.(*Manager).sender+0x116	/home/jacksontj/workspace/golang/src/github.com/jacksontj/promxy/vendor/github.com/prometheus/prometheus/discovery/manager.go:234

21110 @ 0x4374c0 0x405f77 0x405c7b 0x20cc731 0x467571
#	0x20cc730	github.com/jacksontj/promxy/pkg/servergroup.(*ServerGroup).Sync+0x16b0	/home/jacksontj/workspace/golang/src/github.com/jacksontj/promxy/pkg/servergroup/servergroup.go:103

21110 @ 0x4374c0 0x405f77 0x405c7b 0x9ba052 0x467571
#	0x9ba051	github.com/prometheus/prometheus/discovery.(*Manager).Run+0x71	/home/jacksontj/workspace/golang/src/github.com/jacksontj/promxy/vendor/github.com/prometheus/prometheus/discovery/manager.go:142
...

To reproduce start a Promxy instance with the default configuration from the repository and send some SIGHUPs.

---

I found two issues :

• oldState ServerGroups are not cancelled if appenders didn't change
• Prometheus discovery target managers never close SyncCh blocking the ServerGroup Sync() goroutine forever

[camathieu]

ping @jacksontj

[jacksontj]

@camathieu thanks for the contribution! This looks good, but there was a problem with CI. I have pushed up a fix for that, so I'll need a rebase of this so CI can pass (so we can merge!)

[jacksontj]

@camathieu A friendly reminder here -- this should work after a rebase (to include the fix for CI) :)

[jacksontj]

I have rebased this commit into a new PR (#551) -- so closing this one out.