webhooks: add adcp.webhooks.deliver() helper to eliminate seller boilerplate

[bokelley]

Context

Round-4 webhooks DX exploration (PR #205) found that every seller implementing webhook emission rewrites ~30 lines of boilerplate:

1. Read push_notification_config / reporting_webhook from the request
2. Build the payload via adcp.webhooks.create_mcp_webhook_payload(...) or create_a2a_webhook_payload(...) (branch on transport)
3. Sign with adcp.webhooks.get_adcp_signed_headers_for_webhook(...)
4. POST via httpx with json=payload, headers=signed
5. Echo PushNotificationConfig.token into the payload for buyer-side auth
6. Handle authentication.schemes[] (Bearer, HMAC, HMAC+timestamp)
7. Handle retries with byte-identical re-POST

The signer's serialization-format mismatch with json= was already silent-failing across SDKs before PR #205 fixed it. That bug lived in that boilerplate — wherever the seam is between "build payload" and "POST payload," there's room for drift. Collapsing the seam kills the class of bug.

Proposal

async def deliver(
    config: PushNotificationConfig | ReportingWebhook,
    payload: dict | AdCPBaseModel,
    *,
    client: httpx.AsyncClient | None = None,
    secret: str | None = None,  # falls through to config.authentication
) -> httpx.Response:
    \"\"\"One-shot webhook dispatch. Handles signing, token echo, transport
    branching, auth scheme selection, and compact-separator JSON serialization.
    Returns the response for status inspection.\"\"\"

The helper collapses steps 3-6 (and owns step 2 as a convenience) into one call, routes the right payload builder based on the transport hint in config, and signs against the same bytes it POSTs — making the signer/wire-format drift structurally impossible.

Acceptance

• New adcp.webhooks.deliver() function + tests covering: happy path (MCP + A2A), HMAC auth, Bearer auth, token echo from PushNotificationConfig.token, retry idempotency (re-POST produces identical bytes).
• Seller skill's "Emitting Webhooks" section switches to await deliver(config, payload) as the primary pattern. The manual 6-step path stays documented for advanced users who need to customize.

Priority

4.1 — DX improvement, not a bug fix. The fix for the signer/wire-format drift landed in 4.0 (PR #205); the helper would have made it harder for that bug to exist in the first place.

Surfaced during round-4 DX exploration: #205