[Cosmos] Revisit the encoding limitation of the newer version of cosmos

[Olshansk]

Objective

Understand what limitations the newer version of Cosmos imposes on us and adapt appropriately.

Origin Document

Goals

• Read, learn and understand the direction of the Cosmos SDK when it comes protobuf encoding
• Follow best practices in how we are designing the structures (i.e. onchain state) and interfaces (gRPC queries & transactions) in poktroll
• Prepare for the v0.52, v0.2 and beyond

Deliverables

• Review the Zero Copy Encoding document from the Cosmos SDK:
• API (gRPC) Quries & Transactions
  • Decide if we should use the same (i.e. current implementation) structures in APIs or different (i.e. new implementation)
  • Reach out to the noble team and see why this PR for map support in queries was never merged
  • Search for [Cosmos] Revisit the encoding limitation of the newer version of cosmos #1033 in the codebase and tend to all TODOs
• Structures - Evaluate how we need to change any of our onchain structures to comply with:
  • Floats not being a supported scalar
  • Maps not being supported
  • Limitation of not having new fields

    The biggest change is that it will be invalid to add a new field to an existing message and a breaking change detector will need to be created which augments [buf breaking](https://docs.buf.build/breaking/overview) to detect this.
    
    The reasons for this are two-fold:
    
    1) from an API compatibility perspective, adding a new field to an existing message is actually a state machine breaking change which in [ADR 020](https://docs.cosmos.network/v0.52/build/architecture/adr-020-protobuf-transaction-encoding) required us to add an unknown field detector. Furthermore, in [ADR 054](https://docs.cosmos.network/v0.52/build/architecture/adr-054-semver-compatible-modules) this "feature" of protobuf poses one of the biggest problems for correct forward compatibility between different versions of the same module. 2) not allowing new fields in existing messages makes the generated code in languages like Rust (which is currently our highest priority target), much simpler and more performant because we can assume a fixed size struct gets allocated. If new fields can be added to existing messages, we need to encode the number of fields into the message and then do runtime checks. So this both increases memory layers and requires another layout of indirection. With the encoding proposed below, "plain old Rust structs" (used with some special field types) can be used.
    
    Instead of adding new fields to existing messages, APIs can add new messages to existing packages or create new packages with new versions of the messages. Also, we are not restricting the addition of cases to oneofs or values to enums. All of these cases are easier to detect at runtime with standard switch statements than the addition of new fields.
    

Non-goals / Non-deliverables

• Forking the Cosmos SDK
• Creating our own encoding rules / best practices
• Trying to make the Cosmos SDK change based on our own opinions

General deliverables

• Comments: Add/update TODOs and comments alongside the source code so it is easier to follow.
• Testing: Add new tests (unit and/or E2E) to the test suite.
• Makefile: Add new targets to the Makefile to make the new functionality easier to use.
• Documentation: Update architectural or development READMEs; use mermaid diagrams where appropriate.

---

Creator: @Olshansk
Co-Owners: @red-0ne @bryanchriswhite

[Olshansk]

For reference, here is the error that forced us to move rev_share from float to uint64: https://github.com/pokt-network/poktroll/actions/runs/12832899448/job/35787150432?pr=1028