With the high penetration of distributed energy resources and the widespread integration of stochastic loads, the operation of distribution networks exhibits significant temporal coupling and uncertainty. Existing research is often limited to static hosting capacity assessment and feasible operating region verification, while lacking effective methods to quantify and optimize the overall hosting region of the system. To address this issue, a bi-level energy storage planning method for enhancing the hosting region of distribution networks is proposed. First, a distribution network hosting region model is constructed by integrating hosting capacity requirements with feasible operating state space. Second, a dimension-reduction projection-based hosting region volume metric and sensitivity index are developed to quantify the overall size of the feasible operating region under multi-period coupling and to evaluate the marginal benefits of energy storage deployment. On this basis, a bi-level planning model is established aiming at maximizing the hosting region volume while minimizing the overall economic costs. The upper level problem employs an improved particle swarm optimization (IPSO) algorithm to determine the optimal energy storage capacity allocation, while the lower level utilizes a gated recurrent unit-proximal policy optimization (GRU-PPO) framework to solve the dynamic hosting region. Case study results demonstrate that the proposed method accurately characterizes the structural features of the feasible operating region and significantly enhances the system's hosting capacity under the dual uncertainties of source and load.