Currently the Vukan backend will allocate one descriptor pool per render pass: https://github.com/flutter/engine/blob/main/impeller/renderer/backend/vulkan/descriptor_pool_vk.h#L35

Right now we track the descriptor pool and de-allocate it with the rest of the Vulkan tracked objects: https://github.com/flutter/engine/blob/main/impeller/renderer/backend/vulkan/command_encoder_vk.cc#L78 . Instead we could use a similar pattern to cmd pools and reset it.

This should be simpler than with the Cmd pools, but we have the caveat that we need to track the descriptor counts per pool.