Refactor Driver::runInternal (#10906)

Summary:
This PR is part of #10888

The `Driver::runInternal()` function has become too long and too hard to read. This PR refactors it with two changes:

1. Refactor the operator timing code:
    Previously, a DeltaCpuWallTimer needs to be created at the beginning of
    each operator call block. This part of code is usually greater than 4
    lines, making the Driver::runInternal() function too long and hard to
    read. This commit extracts the common code into withDeltaCpuWallTimer()
    function, such that measuring the operator time is easy and clean.

2. Extract common functions so that readers can focus more on the real driver logic.

Pull Request resolved: #10906

Reviewed By: xiaoxmeng

Differential Revision: D62995248

Pulled By: Yuhta

fbshipit-source-id: 88ea016b8c68adf5349fc6695341c79b5457f4cf