[pt2][inductor] turn off cache search by default (#95662)
Summary: set `search_autotune_cache=False` by default due to inductor compilation regression on HF models, while working on reducing overhead
Test Plan: CI
Differential Revision: D43641286
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95662
Approved by: https://github.com/ngimel, https://github.com/jansel
