A2A agent url is not checked for embedded userinfo, unlike the Vertex and Azure api_base

What

A2aAgent.url is accepted and dialed as configured. The Vertex and Azure bridges both reject an api_base that embeds userinfo (https://user:pass@host, #390 / #391) because a URL-embedded credential is not a credential the gateway holds or can rotate, and it leaks into anything that renders the URL. The A2A path has no equivalent check.

Why it matters

• The credential travels on every dispatch to that agent, outside the auth_type / secret mechanism the resource models for exactly this purpose.
• A2aUpstream's Debug and the agent-card warning path both render the URL, so it reaches logs.

Neither is a regression — the URL has always been used verbatim — but the two sibling bridges treat this as a config error and A2A does not, which is the kind of asymmetry that survives because nobody is looking at both at once.

Suggested fix

Reject at resource-resolution time (upstream_from_a2a_agent, or wherever the A2A agent resource is validated), with the same shape of error the Vertex and Azure bridges return, so the operator sees an actionable message instead of a silently credential-bearing URL.

Found by the independent pre-merge audit of api7/aisix#955, which noted it as pre-existing and out of that PR's scope.

Start at resource resolution in upstream_from_a2a_agent and compare its URL handling with the Vertex and Azure api_base validation for issues #390 and #391. The work is done when A2A agent URLs containing embedded userinfo are rejected with an equivalent actionable configuration error before dialing or rendering the URL.