test(rpc): equivalence of execution vs header randomness providers (split from #367)

[nekomoto911]

Summary

Pin the byte-equality of the two RandomnessByHeightProvider implementations with a unit-level equivalence test. Single commit factored out of #367 — not dependent on Gravity Alpha.

Why split out of #367

The randomness_by_height precompile is registered in two places with intentionally distinct data sources:

• pipe ExecutionRandomnessProvider (crates/pipe-exec-layer-ext-v2/execute/src/randomness_precompile.rs) reads the in-flight OrderedBlock.prev_randao for the current block, parent_header.mix_hash() for the parent, and a GravityStorage-backed fallback for older heights
• RPC HeaderRandomnessProvider (crates/rpc/rpc/src/eth/helpers/call.rs) reads HeaderProvider::header_by_number(h).mix_hash for past heights and uses the EVM env's prev_randao (passed as current_randomness) for the simulated block

The two diverge on data source by necessity (the in-flight block is not persisted while it is executing) but the shared trait RandomnessByHeightProvider requires they produce byte-identical RandomnessByHeightLookups on the same (block_number, query_height) input for RPC replay (debug_trace* / trace_*) to stay byte-equal canonical on any historical block calling 0x...625f5002.

Same class of pipe/RPC gap as the BLS RPC-registration bug — pinning it with a unit test is independent of Alpha.

What

Adds execution_provider_and_header_provider_agree_byte_for_byte in crates/rpc/rpc/src/eth/helpers/call.rs#tests. Feeds both providers the same in-memory dataset and asserts RandomnessByHeightLookup byte-equality across 11 query_height boundaries: 0, 1, current / 2 (a height with no header), current - window - 1, current - window, current - window + 1, current - 2, current - 1 (parent), current, current + 1, u64::MAX. The fallback (pipe GravityStorage) vs header_by_number(h).mix_hash (RPC) data-source agreement is a separate engine-API ingestion invariant, not under test here.

Adds reth-pipe-exec-layer-ext-v2.workspace = true to crates/rpc/rpc/Cargo.toml's [dev-dependencies] so both provider types are in scope from the #[cfg(test)] module (HeaderRandomnessProvider is local to reth-rpc; ExecutionRandomnessProvider lives in pipe-exec-layer).

Test plan

• Local cargo check -p reth-rpc --tests
• CI: cargo nextest run -p reth-rpc execution_provider_and_header_provider_agree_byte_for_byte

Compatibility

Test-only + dev-dep. No prod code changes.

Coordination

• vs feat(hardfork): gas-exempt system txs + zero SYSTEM_CALLER at Alpha #367: feat(hardfork): gas-exempt system txs + zero SYSTEM_CALLER at Alpha #367 contains the same commit (byte-identical patch from cherry-pick). After this merges, feat(hardfork): gas-exempt system txs + zero SYSTEM_CALLER at Alpha #367 needs `git rebase upstream/main` to drop it (handled by git's empty-commit detection).
• vs the BLS RPC split PR: that PR also adds `reth-pipe-exec-layer-ext-v2.workspace = true` to the same `[dev-dependencies]` block. Whichever merges second needs a one-line rebase to drop the duplicate add.