A Few Expert Queries Suffices for Sample-Efficient RL with Resets and Linear Value Approximation