feat: Draft: Add link-awareness

[mterhar]

Which problem is this PR solving?

• How should we handle linked traces? #516

Short description of the changes

• Adds the ability for one refinery node to ask another one if it should keep a trace
• Adds the the decision-by-link from a root span back to a preceding trace
• Adds a configuration called LinkStrategy that can be enabled to allow link-awareness
• Caches it's answer in the cuckoo to limit additional api traffic

Things to wrap up:

• marshal the response into JSON to be more deliberate about how it's handled
• see how it handles links to links

Things it doesn't do:

• Doesn't look at multiple links on the root span
• Doesn't look at additional spans farther down the trace. Just the root span linking back to any span on the other trace.
• Doesn't push a decision up if a the predecessor trace is still pending and the linked trace matches an error rule.
• Doesn't lock on the API call so if a bunch of linked traces arrive at once, it'll make lots of API calls at the same time
• Doesn't allow the link to be considered among other things. Right now it's all-or-nothing.
• Doesn't allow link handling to be done different ways depending on the environment
• Doesn't share the decisions into a shared storage (like the redis thingee)

This is up for discussion. Comments encouraged.

[bixu]

We've had folks at $job ask about this when we suggest they break up "megatraces" into smaller ones.