[SR] Factor logic that determines managed tensors out of MemoryPlanner (#68295)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68295
There's no reason we can't figure out what tensors we need to manage at model load time. It's also useful to have the set of ranges available at load time for integrating the ranges algorithm introduced in the previous diff.
Test Plan: `buck test caffe2/benchmarks/static_runtime/...`
Reviewed By: hlu1
Differential Revision: D32400593
fbshipit-source-id: 0466b2641166ddc9c14f72774f4ba151407be400
