Synthetic regression fixture for automatic x402 retry

[echowang1]

Concrete synthetic assertions for x402-openai-python automatic payment handling:

• An initial 402 with amount at or below max_amount selects the matching network wallet, signs once, and retries once.
• An amount above max_amount fails before any wallet signature is produced.
• Network and scheme policies select the intended wallet and payment requirement when alternatives are present.
• A successful paid retry returns the OpenAI response without another payment attempt.
• A second 402 on the paid retry does not recurse into another sign-and-pay cycle.

Gauntlet is a local-only V0.1 assertion harness, not production payment security.

Would this synthetic retry-and-policy fixture be useful for the QNTX x402 OpenAI Python client?