Parallelizing Resize op

[duli2012]

Description: Describe your changes.
Make bilinear and trilinear mode of Resize parallel

Motivation and Context

• Why is this change required? What problem does it solve?
  This change makes the resize op 3 times faster for a product model.

Credit to @tlh20

[duli2012]

Currently, it's using ThreadPool::TrySimpleParallelFor API. We'd like to use ThreadPool::TryParallelFor which is more robust and effcient. However, the latter has unaddressed issues related Eigan cost model, which we'll revisit later.

[tlh20]

Couple of minor formatting glitches. Only substantive comment is if we need to fall back to sequential execution on any cases.