transcoder: Move create callback into benchmark state loop

When the google benchmark iterator loop executes
more than once, using the same callback results
in abort. This is reproducible when the transcoding
operation is of the order of 100s of milliseconds
and so the benchmark attempts to run more
iterations. To fix the issue, moved creating
callback into benchmark state loop.

Bug: 175889103
Test: ./MediaTranscoderBenchmark

Change-Id: I20a435a93ed8ccb3c8fd3d27053b20b54d9d6e26