server: need more scalable protocol
Discussion in #17539 (closed) and related MRs/issues identified that the current monolithic codemodel query response is not scalable. It can grow arbitrarily large, and any new information added to it (e.g. backtraces) make the problem larger.
Using granular queries could solve the scalability problem but creates latency problems.
I'm opening this issue to serve as a dedicated place for discussion of this problem.