[Operators] preliminary symmetric weight quantization

[Aalanli]

• Added preliminary symmetric weight quantization in tasks
• Changed Operator._run to Operator.symbolic_run, such that construction of the operator does not automatically run imperative_run, but symbolic_run. This is so that constant folding does not happen with barrier op, and leads to the intended behaviour with quantization layers. This should be fine to do since all imperative calls calls op.get_output(...), anyways, which calls imperative_run

[yaoyaoding]

Thanks @Aalanli !

I left some comments, and made some update directly on this PR.

I refactored a little bit on the run, symbolic_run and imperative_run.

[Aalanli]

Thanks for the changes, makes sense.

[yaoyaoding]

Thanks @Aalanli !