[dynamo benchmarks] Fix inference benchmark runs (#98248)
Update flags for dynamo inference benchmark runs. Add flag to not compute regressions/metric graphs (useful if there aren't previous runs to compare with).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98248
Approved by: https://github.com/shunting314
