[ dev ] Adding `universal_research` tier

[toilaluan]

Overview:

Adding new protocol that miner returns a text. It can be used directly with any LLM models, that's why we call it "universal".

Validator <--text--> miner <--condensed_text--> validator

Reason:

• Reducing token length by using gist token or pruned kv-cache is only usable with a specific model and not easily to use with other LLM providers for now.

Target

• Implement baseline for miner: receive a long context and return compressed text
• Implement validator logic for validating compressed text based on Question Answering
• Allocate 10% incentive for this tier. It means current research tier incentive is 90%. We need to maintain both of them.