Electric mobility service platforms are increasingly expected to coordinate mobility, logistics, and energy services within the same fleet system. This integration gives rise to a stochastic fleet sizing and scheduling problem in which a platform must determine its platform-owned electric vehicle (EV) fleet while using flexible EV supply, such as temporary or part-time EV resources that can enter the system at available locations and times and leave after their service commitments, to absorb uncertain demand and travel time variation. The resulting operations require EVs to be allocated across ride-hailing, on-time delivery, and vehicle-to-grid (V2G) services under time window, capacity, battery, charging, station-capacity, rest, and fleet-availability constraints. This study develops a stochastic optimization framework for sizing and operating multi-service electric vehicle fleets. The problem is formulated on a time-expanded action network, where each arc represents a feasible vehicle action, including service fulfillment, repositioning, charging, V2G discharging, waiting, and rest. Ride-hailing, delivery, and mixed service sequences are represented by preprocessed service-pattern actions, which embed pickup-completion precedence, ride-pooling, delivery consolidation, capacity feasibility, delay, detour, and energy-consumption information. A scenario-based stochastic formulation is used to capture uncertain demand and travel times, and jointly optimizes platform-owned fleet sizing, request acceptance, flexible-supply activation, vehicle scheduling, charging, and V2G operations. To solve the resulting large-scale mixed-integer program, we develop a customized branch-and-Benders-cut algorithm and a column generation-based matheuristic. Numerical experiments demonstrate the effectiveness and computational efficiency of the proposed algorithms, and show that stochastic integrated planning improves platform profitability.