fix: remove stale NonceTooLow txs from retry loop

[qj0r9j0vc2]

Summary

• Track permanently-skipped (NonceTooLow) transaction hashes in ExecutionResult
• Remove them from the RPC pending map after block execution
• Prevents infinite retry loop where stale transactions cycle through consensus and execution repeatedly

Problem

After block execution, retry_pending() re-sends all unexecuted pending transactions to the Worker channel for rebatching. However, NonceTooLow transactions can never succeed (their nonce has already been used), so they cycle forever: Worker → Batch → Consensus → Execution (NonceTooLow) → retry → Worker → ...

This produces blocks where all transactions are skipped, wasting consensus bandwidth and producing empty blocks.

Changes

• ExecutionResult.skipped_tx_hashes — new field tracking NonceTooLow tx hashes
• engine.rs — collects skipped hashes during process_transactions
• BlockExecutionResult — propagates skipped hashes through execution bridge
• node.rs — removes skipped txs from pending map and excludes from retry

Test plan

• cargo check passes
• cargo test -p cipherbft-execution -p cipherd passes
• Verify under load test that NonceTooLow transactions no longer cycle