feat(near): added sui support in mpc prover

[frolvanya]

Summary

Adds the NEAR-side prover path for Sui, so InitTransfer / FinTransfer /
DeployToken / LogMetadata events emitted by the Sui locker can be proven
back to NEAR. This was the last code blocker listed in sui/README.md;
everything remaining after this is DAO calls and deployment.

Mirrors the Aptos rollout.

What's included

• near/omni-types/src/sui/events.rs — parsers for the four bridge
  events, plus parse_sui_event / parse_sui_proof dispatch.
• near/omni-types/src/sui/bcs.rs — minimal BCS reader (no new deps).
• MpcFinality::Sui(SuiFinality::Checkpointed) and the ChainKind::Sui
  arm in mpc-omni-prover (request_to_chain_kind,
  request_matches_finality, verify_callback, seeded in init).
• near-mpc-sdk rev bump dcc9e042 -> 70c40b0f (merge commit of
  feat(foreign-tx): sui support in contract interface near/mpc#3754, which adds the Sui contract-interface DTOs). The old pin has
  no Sui types at all. Verified no breaking changes for EVM / Starknet /
  Aptos.

Two things worth reviewing closely

1. BCS, not JSON. Aptos delivers AptosEvent.data as a JSON rendering;
Sui delivers SuiEvent.bcs as canonical BCS bytes. Parsing is therefore
positional and must match the Move struct declaration field-for-field — note
the Sui events carry an extra coin_type: String between token_address and
the numeric fields, which Aptos does not have.

Two guards make a layout mistake loud instead of silent:

• Reader::finish() rejects trailing bytes, so a wrong field order fails
  rather than producing a plausible-looking wrong value;
• ULEB128 lengths are canonical-only (BCS uses ULEB128 where Borsh uses a
  fixed u32 — easy to get wrong).

2. Emitter comes from type_tag, not event.package_id. A Move type
keeps its defining package id across package upgrades, while package_id
becomes the upgraded package's id. Only the former stays equal to the factory
address registered on NEAR, so this survives a bridge package upgrade. Same
rule as the Aptos type-tag-address convention.

One extra check Aptos can't do

Sui coins are types, so every event carries both the 32-byte wire id and the
coin_type string it derives from. The parsers assert
keccak256(coin_type) == token_address, binding the two representations.

Tests

113 passing (+25): 77 in omni-types, 36 in mpc-omni-prover.

BCS fixtures are generated by an independent Python reference encoder, so the
tests don't just assert the Rust against itself. Coverage: all four events,
tag dispatch, proof-kind/tag mismatch, the keccak binding (flip one byte of
the token id), trailing and truncated payloads, short-form addresses,
malformed tags, non-canonical ULEB128, non-UTF-8 strings, and the
prover-level dispatch/finality/roundtrip cases mirroring Aptos.

Verified: cargo test, make clippy-near (pedantic, -D warnings),
make fmt-near, and cargo check --target wasm32-unknown-unknown.

[frolvanya]

@claude review

[claude]

Pull request overview

Adds the NEAR-side MPC-prover path for Sui, so InitTransfer / FinTransfer / DeployToken / LogMetadata events emitted by the Sui locker can be proven back to NEAR. This is the Sui analogue of the Aptos rollout (#626 / #629) and is stacked on feat/sui (which already carries ChainKind::Sui / OmniAddress::Sui). The one structural difference from Aptos is that the MPC layer delivers Sui events as canonical BCS bytes (SuiEvent.bcs) rather than a JSON rendering, so parsing is positional and must mirror the Move #[event] structs field-for-field.

Changes:

• New near/omni-types/src/sui/{bcs.rs,events.rs,mod.rs}: a minimal positional BCS reader and four event parsers + parse_sui_event / parse_sui_proof dispatch.
• mpc-omni-prover: MpcFinality::Sui(SuiFinality::Checkpointed) seeded in init, plus ChainKind::Sui arms in request_to_chain_kind, request_matches_finality, and the result-parse dispatch (parse_sui_result).
• MpcFinality::Sui appended to the enum in mpc_types.rs.
• near-mpc-sdk git rev bump dcc9e042 → 70c40b0f (adds the Sui contract-interface DTOs).
• Doc updates (near/CLAUDE.md, sui/README.md).

Reviewed changes

Per-file summary

File	Description
near/omni-types/src/sui/bcs.rs	New positional BCS reader (u8/u64/u128, address, vector, String, Option, ULEB128) with trailing-byte / canonical-ULEB / UTF-8 guards
near/omni-types/src/sui/events.rs	Four BCS event parsers, emitter-from-type_tag derivation, keccak256(coin_type)==token_address binding, ProofKind/type_tag dispatch
near/omni-types/src/sui/mod.rs	Module wiring
near/omni-types/src/lib.rs	pub mod sui;
near/omni-types/src/mpc_types.rs	MpcFinality::Sui(SuiFinality) variant appended
near/omni-prover/mpc-omni-prover/src/lib.rs	Sui finality seed + ChainKind::Sui dispatch to parse_sui_result
near/omni-prover/mpc-omni-prover/src/tests.rs	Sui prover-level tests (dispatch, finality, roundtrip, rejects)
near/Cargo.toml / Cargo.lock	near-mpc-sdk rev bump
near/CLAUDE.md, sui/README.md	Docs

Findings

Blocking: none identified.

I traced all four BCS test fixtures byte-for-byte and each decodes exactly as the parser reads it and as the doc-comment layout claims (e.g. InitTransfer = sender(32) | token(32) | uleb(0x4e=78) coin_type | nonce=7 LE | amount=1000 | fee=10 | native_fee=5 | uleb(0x13) "near:frolik.testnet" | uleb(0x00) message). The guards are sound: finish() rejects trailing bytes, uleb128 rejects non-canonical/over-u32 lengths, strings enforce UTF-8, Option rejects tags >1, origin_chain goes through TryFrom<u8>, and every parser both checks the type_tag suffix and binds keccak256(coin_type) == token_address. Prover wiring mirrors Aptos completely (init map, request_to_chain_kind, request_matches_finality, parse dispatch), and appending MpcFinality::Sui is borsh-safe (no existing discriminant renumbered).

Non-blocking (verify before merge — no automated gate covers this):

• near/omni-types/src/sui/events.rs:653,694,726,753 — The positional BCS layouts cannot be verified against the source of truth in this review. The actual Sui Move #[event] structs live in sui/sources/omni_bridge.move on the base branch feat/sui, which isn't in this diff, and per §8 there is no Sui CI workflow and no cross-chain payload gate — the round-trip tests use fixtures from an independent Python encoder, so they prove the parser agrees with that encoder, not with the Move contract. A single mis-ordered/mis-typed field would compile, pass all tests here, and silently brick the Sui lane. Please diff each parser's field order/types against the Move #[event] declarations one field at a time before merging.
• near/omni-types/src/sui/events.rs:704 — Two specific spots to confirm against the Move source, since they differ from the Aptos mirror: (1) coin_type: String is read immediately after token_address in all four events (the new field), and (2) FinTransfer.message is read as a plain vector<u8> here, whereas Aptos declares it Option<vector<u8>> (aptos/events.rs:191). If the Sui struct actually uses Option<vector<u8>>, a Some(..) message would misalign and fail finish() (a None/empty message would still pass), so this only surfaces on a message-bearing FinTransfer — worth an explicit confirmation.

✅ Approved