fix: widen happy-eyeballs connect window for ipfs proxy

[erubboli]

Problem

Token icons kept flaking even after the gateway failover and retry fixes. In-container tracing caught connect ETIMEDOUT failures consistently at ~505 ms — far below the 12 s gateway budget. That signature is Node's happy-eyeballs per-address attempt timeout (250 ms default) firing twice: the RTT to the CDN edge from this network is ~300 ms, longer than the per-address window, so a share of connects were being aborted mid-handshake even though the destination was healthy. The proxy then returned 502, token metadata resolved to null, and the UI fell back to letter avatars.

Fix

• Raise net.setDefaultAutoSelectFamilyAttemptTimeout to 1000 ms in the ipfs proxy module (server-side). Healthy-but-slow connects now survive; genuinely dead addresses still fall through to the next address family.

Verification

• In-container correlation test: DNS is A-first; failures carried the 2x250 ms signature; with the wider window the same fetch loop succeeds consistently.