e2e: bound per-attempt fetch timeout in spawn waitForReady

The CI flake we kept hitting (opencode serve binds 0.0.0.0:PORT,
prints 'server listening', completes SQLite migration, then the
loopback HTTP probe times out at 300s) was a Bun fetch issue, not an
opencode issue. Bun has a hardcoded ~5 minute fetch timeout that
ignores AbortSignal.timeout values longer than the limit
(https://github.com/oven-sh/bun/issues/16682). The probe's fetch()
calls had no AbortSignal, so a single hung request could hold the
loop for the entire ~5 minute window — exhausting the overall 300s
deadline with curl-fallback diagnostics never firing
(curlEverSucceeded=false, curlLastErr=null in error output).

Fix: bound every fetch attempt at 2s via AbortSignal.timeout so no
single request can starve the deadline. Removed the curl-fallback
path entirely since the root cause was fetch hanging, not fetch being
fundamentally unable to reach the server.