consensus: Frontier Phase F-3 — real parallel apply (replace F-2 shadow)

[satyakwok]

Currently apply_block_pass2 is sequential within a block. F-2 shadow scheduler (gated by SENTRIX_FRONTIER_F2_SHADOW=1) batches transactions to demonstrate the parallel-apply structure without committing state mutations. F-3 replaces F-2's shadow path with a production code path that actually parallelises tx execution while preserving determinism.

Why

Mainnet bt is currently 2-3 s/blk steady-state. Per-block apply profile (gated by SENTRIX_APPLY_PROFILE=1) shows trie update is the dominant chunk at ~340ms per block. Parallel apply across non-conflicting tx batches could amortise that.

Scope

• F-3: real parallel apply across non-conflicting txs in the same block
• F-4 through F-10: progressive enabling on testnet → mainnet
• Cross-validator determinism tests (state_root match across 4 vals after parallel apply)

Constraints

• Per CONTRIBUTING.md, consensus-protocol changes need discussion before submission and bug fixes need a regression test. Each F-N step ships as its own PR with a focused diff.
• ~6-8 weeks calendar including testnet shadow-bake + mainnet activation

Acceptance

• F-3 ships as a separate fork-gated activation height
• Mainnet bt improvement measurable via apply-profile