feat: introduce code for JSON encoding protobuf types

[kruskall]

Introduce code for JSON encoding protobuf types

I tried to keep them close to the original apmevent toModelJSON but it should be fairly contained

Related to #36
Blocked by #41

[axw]

Looks good! I think the CompressionStrategy conversion is wrong, so that's a blocker, but the other things I mentioned are debatable.

While you're here, would you mind changing the port fields in protobuf from uint32 to uint16?

[kruskall]

While you're here, would you mind changing the port fields in protobuf from uint32 to uint16?

Could you clarify this ? Protobuf doesn't have a uint16 AFAIK

[axw]

While you're here, would you mind changing the port fields in protobuf from uint32 to uint16?

Could you clarify this ? Protobuf doesn't have a uint16 AFAIK

Ah right, I had forgotten this. They will be varints, and the number of bytes will depend on the value. Nothing to do here then.

[kruskall]

I've updated the code, here's the benchmarks:

Classic model (model -> modeljson):

BenchmarkAPMEventToJSON/to-json-20                553550              1953 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                558913              1875 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                564428              1898 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                598153              1932 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                626432              1963 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                597972              1942 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                632714              1967 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                504571              2004 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                566557              1898 ns/op            3688 B/op         53 allocs/op
BenchmarkAPMEventToJSON/to-json-20                619671              1960 ns/op            3688 B/op         53 allocs/op

Protobuf (modelpb -> modeljson):

BenchmarkAPMEventToJSON/to-json-20                453487              2596 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                457846              2675 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                430870              2643 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                431421              2605 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                490472              2658 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                474543              2533 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                473373              2783 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                428856              2656 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                463778              2621 ns/op            5096 B/op         52 allocs/op
BenchmarkAPMEventToJSON/to-json-20                418586              2568 ns/op            5096 B/op         52 allocs/op

Notes:

• allocations are mostly the same but memory usage increased: should be addressed by Avoid map allocations when mapping modelpb to modeljson  #48
• the protobuf mapping is slower: while this is true, the JSON encoding logic received some performance improvements in model: introduce separate types for JSON encoding #35. I don't think the "slowdown" here is going to affect us much but we will be able to have more accurate numbers once this is integrated in APM Server

[kruskall]

Pushed a few more fixes I found while integrating this with APM Server (no PR yet). This should be ready for review again.

[axw]

Thank you!

[axw]

Suggested change

	delete(in, k)
	in[sanitizeLabelKey(k)] = v
	if k2 := sanitizeLabelKey(k); k2 != k {
	delete(in, k)
	in[k2] = v
	}

We should expect most keys to not require sanitisation, and optimise for that.