fix(validator): serialize hotkey contract writes to prevent nonce collisions

[jaso0n0818]

Problem

The validator signs contract writes with the same hotkey over two separate substrate connections: the forward loop (contract_client on self.subtensor) and the axon handlers (axon_contract_client on axon_subtensor).

Both call create_signed_extrinsic, which auto-fetches the nonce via AccountNonceApi.account_nonce — the best-block nonce, which does not count pending pool txs. Within the same block window both clients fetch nonce N; one tx lands and the other is banned (error 1012 Transaction is temporarily banned), starving the forward loop and causing delivered swaps to be slashed. Closes #457.

Fix

allways/contract_client.py

Add an optional write_lock: threading.Lock parameter to AllwaysContractClient.__init__. Inside exec_contract_raw, hold the lock across nonce-fetch + submit + inclusion. The pre-flight balance read is intentionally left outside the lock so reads remain parallel.

A _NullContext no-op is used when no write_lock is passed (backward-compatible).

neurons/validator.py

Create one shared self._write_lock = threading.Lock() and pass it to both contract_client and axon_contract_client at construction. Lock order: axon_lock → write_lock (axon writes) and write_lock → substrate_lock (forward writes) — no cycle.

Tests

New tests/test_write_lock_serialization.py (6 tests):

• Two clients sharing the same lock object store the same reference
• Client without write_lock completes exec_contract_raw normally
• write_lock is held when substrate_call(submit_extrinsic) runs
• Balance read (substrate_call for account info) fires before the write_lock is acquired

Closes #457

[jaso0n0818]

Rebased on latest test. Serializes hotkey contract writes to prevent nonce collisions. Ready for review.