varint encoding for trace and span ids is slow

[ethercrow]

I've been staring at profiling data and noticed that putVarInt function is rather prominent there, specifically called when serializing trace and span ids. Since those are randomly generated the probability of them fitting into only a few bytes is vanishingly low, preventing any benefits of using variable length encoding.

What do you think about updating collector.proto to define those as fixed64? Quoting Protobuf Developer Guide:

fixed64 | Always eight bytes. More efficient than uint64 if values are often greater than 2^56.

Or maybe Lightstep already can accept trace data in some other protobuf format that is more efficient?