Add optional AOTrust notarization for agent-executed trades (x402, Ed25519 PDR)

Hi! Great MCP server — we've been running a similar FastMCP production service and know how much care this stack takes.

I'd like to propose an **optional** integration: cryptographic receipts for agent-executed trades.

**The gap:** when an AI agent executes a trade via your MCP server, the agent's principal usually has no tamper-evident, independently-verifiable record of *what* the agent did, *when*, and *under which authorization*. Internal logs work, but they're self-issued — an auditor has to trust the operator.

**What AOTrust adds:** a 239-byte Ed25519-signed receipt (PDR — Provenance Data Record) that proves a specific action happened at a specific time. Receipts are anchored to NEAR blockchain daily, so anyone can verify them offline — no account, no install, no trust in our servers:

```
verify.aotrust.link/?pdr=
```

or programmatically via the public API. The PDR binary spec and an offline Python verifier are open source.

**Proposed change:** add `aotrust-protocol` (on PyPI, MIT) as an *optional* dependency. One line per call site:

```python
from agent_notary import AOTrustTool
receipt = AOTrustTool.run(work_hash=trade_hash) # → Ed25519 PDR, $0.01 x402 (or free tier)
```

- Payment: x402 protocol, $0.01 USDC on Base (EIP-3009), plus a free anonymous tier (5 receipts/IP/24h)
- Works as plain HTTP — no wallet needed on the server side
- Spec + offline verifier + SDK: https://github.com/GitSerge-crypto/aotrust-skills
- Discovery: https://shield.aotrust.link/.well-known/x402.json

Happy to adjust scope (e.g., only for `get_trade_history`-adjacent write ops, or behind a feature flag) — and happy to answer any questions about the trust model. We don't need to trust the notary: spec + offline verifier + daily NEAR anchor means anyone can check receipts independently.