Make test test_local_deployment use chain fork instead of the real chain

[gabrielfior]

The test test_local_deployment uses Gnosis Mainnet in its attempt to run the agent - basically it triggers agent.run (without placing bets), but it does try to redeem positions.
This is interfering with the operation of the deployed agent - see errors e.g. here
https://github.com/gnosis/prediction-market-agent-tooling/actions/runs/10479070131/job/29023879784#step:7:262
and here
https://github.com/gnosis/prediction-market-agent-tooling/actions/runs/10478961453/job/29023515375#step:7:223

Note that the position_id as well as the index on the logs increases, thus if you re-try long enough the tests will eventually pass.

The test would be more robust if executed on a local fork (using the existing pytest fixture local_web3), since all positions exist there as well.