Add most ops (parity w/ fp32) (#1797)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/1797

Add all the ops in a single diff because since we are touching xnnpack, each one runs 250K jobs and killing infra in the process. Two ops are not in this, linear and sdpa.

Reviewed By: mcr229

Differential Revision: D53248927

fbshipit-source-id: 4fb3fdc4e358d29cce7c49af5c9d851470018cdb