Add span deduper to support zipkin client/server kind spans

[annanay25]

Testing:

curl -X POST http://localhost:9411 -H 'Content-Type: application/json' -d '[{
 "id": "12345",
 "traceId": "0123456789abcdbb",
 "timestamp": 1608239395286553,
 "duration": 100000,
 "name": "span from bash!",
 "kind": "SERVER",
 "parentId": "6b221d5bc9e6496c"
 "tags": {
    "http.method": "GET",
    "http.path": "/api"
  },
  "localEndpoint": {
    "serviceName": "shell script"
  }
},
{
 "id": "12345",
 "traceId": "0123456789abcdbb",
 "timestamp": 1608239395286533,
 "duration": 100000,
 "name": "span from bash!",
 "kind": "CLIENT",
 "tags": {
    "http.method": "GET",
    "http.path": "/api"
  },
  "localEndpoint": {
    "serviceName": "shell script"

  }
}]'

Signed-off-by: Annanay annanayagarwal@gmail.com

Benchmark results

$ go test -bench=.
goos: darwin
goarch: amd64
pkg: github.com/grafana/tempo/modules/frontend
cpu: Intel(R) Core(TM) i5-8279U CPU @ 2.40GHz
BenchmarkDeduper100-8      	   89684	     13172 ns/op
BenchmarkDeduper1000-8     	    7178	    163071 ns/op
BenchmarkDeduper10000-8    	     738	   1567722 ns/op
BenchmarkDeduper100000-8   	      57	  19311762 ns/op
PASS
ok  	github.com/grafana/tempo/modules/frontend	6.102s

I think these are ok to proceed.

What this PR does:
Adds a span deduper middleware that allows multiple spans of the same ID but different spanKind (client/server).

This is supported in Zipkin.

Which issue(s) this PR fixes:
Fixes #682

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[joe-elliott]

Two comments but otherwise looks good so far. Besides the comments two other things that would be nice to see:

• a comment to indicate that this is being done to support zipkin
• benchmarking of various trace sizes: 100, 1k, 10k, 100k spans

I'm a little nervous about the impact on correctness but with @kvrhdn 's improvements to the vulture I think we will feel better.

[annanay25]

Just need to add benchmarks now.

[joe-elliott]

Some comments. I think you need to make a change to your benchmark but I don't expect it to appreciably change the results.

Agree if the results stay roughly the same that we can just always run the deduper.