test(e2e): bump OC long-running sendPrompt to 600s

Round 5 still showed long-running OC test failing at exactly 300s for
one sendPrompt call. The test does 27 turns sequentially with historian
subagent publication, and one of those turns can take 5+ minutes when
the subagent process cold-starts under CI runner contention.

Bumped sendPrompt budget for the long-running test from 300s → 600s
(matching the per-test it() budget). All other OC tests pass at the
default 180s harness budget.

No production code changes.