bug: a2a isAlive treats HTTP 500 as alive

[Ridgeio]

Summary

A2ATransport.isAlive() in src/a2a-transport.ts treats a non-2xx agent-card response as a reason to fall back to a bare HEAD, then returns true for any resolved HEAD response regardless of status. An A2A endpoint returning persistent 500 is therefore reported as alive, while network-unreachable is reported as dead.

Evidence

• src/a2a-transport.ts:29 — HEAD path does not check resp.ok / status code.

Impact

• swarm members and the lazy-reap path (new in PR Lazy reap on join conflict + swarm reap CLI #1) both treat a broken-but-responding A2A agent as alive, so swarm join <name> for a dead-but-5xx A2A agent is blocked even though delivery will fail.
• Real-world messages fail silently downstream instead of being cleanly rejected at lookup.

Fix sketch

In isAlive(), require resp.ok (or at least non-5xx) on both the agent-card and HEAD fallback paths before returning true.

Context

Surfaced by codex challenge during review of PR #1 (lazy-reap). Out of scope for that PR — filing here per Lead.