Raise default dial timeout 500ms -> 2s

A 500ms per-host connect budget sometimes trips on a slow initial connect (cold DNS, TLS-fronting LB, distant region) before the node can answer. 1s still fails over off a dead node quickly while absorbing that cold start. ENTIRE_CONNECT_TIMEOUT_SECONDS remains the escape hatch for slower links.

Hoping this will make https://entireio.slack.com/archives/C09V780RGGN/p1780325899840859 less likely.