Read via FileAdapter when loading files in torch if not flatbuffer (#84028)
Summary: This will optimize memory usage at the small cost of loading time when loading mobile models restoring the behavior before D36926217 (https://github.com/pytorch/pytorch/commit/fed12ff680813c0fab7dba7232f6b4cd8b33b8d3).
Test Plan: Signals
Reviewed By: qihqi
Differential Revision: D38998858
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84028
Approved by: https://github.com/qihqi, https://github.com/cccclai