fix: set signatureChainId in recover_user_from_user_signed_action

[massimiliano1991]

Summary

recover_user_from_user_signed_action is the inverse of sign_user_signed_action, but the two are asymmetric:

• sign_user_signed_action sets both signatureChainId and hyperliquidChain on the action before building the EIP-712 payload.
• recover_user_from_user_signed_action sets only hyperliquidChain.

Because user_signed_payload reads action["signatureChainId"], recovering the signer from a freshly reconstructed action — the normal case for a verifier that never called sign_user_signed_action on that dict — raises KeyError: 'signatureChainId'.

Reproduction

from hyperliquid.utils.signing import (
    sign_usd_transfer_action, recover_user_from_user_signed_action, USD_SEND_SIGN_TYPES,
)
import eth_account

wallet = eth_account.Account.from_key("0x" + "01" * 32)  # any test key
action = {"destination": "0x0000000000000000000000000000000000000001", "amount": "1", "time": 1677777606040}

sig = sign_usd_transfer_action(wallet, dict(action), True)
recover_user_from_user_signed_action(dict(action), sig, USD_SEND_SIGN_TYPES, "HyperliquidTransaction:UsdSend", True)
# -> KeyError: 'signatureChainId'

Fix

Set signatureChainId in recover_user_from_user_signed_action symmetrically with sign_user_signed_action. Added a sign → recover round-trip regression test (mainnet + testnet) that asserts the recovered address equals the signer — the function previously had no test coverage, which is why this slipped through.

pytest tests/signing_test.py passes (14 tests).

Note on #267

This is orthogonal to #267 (which makes signatureChainId configurable in sign_user_signed_action). The fix here matches current master, where sign_user_signed_action hardcodes 0x66eee. If #267 lands, the recover side should stay in sync — ideally both sign and recover would share a single helper for the default so they can't drift apart again.

[koriyoshi2041]

Verified at a08ae6e761. A freshly reconstructed USD-send action fails with KeyError: 'signatureChainId' on master, while this restores the same EIP-712 domain input used by sign_user_signed_action; mainnet and testnet round trips recover the signer correctly. The full local suite passes (37/37) on Python 3.11, and git diff --check is clean.

The hardcoded value deliberately matches current master. If #267 later makes it configurable, both sign and recover paths will need to move together.

[massimiliano1991]

Thanks for the thorough verification — reproducing the KeyError on master and running the round trips on both chains is more than most PRs get. Agreed on the #267 coupling: the hardcoded value deliberately mirrors current master, and if it becomes configurable the sign and recover paths have to move together. Happy to update this PR if that lands first.