[NFC] Parallelize the actual inlining part of the Inlining pass

[kripken]

We already parallelized collecting info about functions and finding
inlining opportunities, but the actual inlining - copying the code into
the target function - was done sequentially. It turns out that a lot of
work happens there: this PR makes the pass over 2x faster.

Details:

• Move nameHint to InliningAction, so it is there when we apply the actions.
• Add a DoInlining internal pass which calls doInlining().
• Refactor OptUtils a little to make it easy to run DoInlining before opts.
• Also remove the return value of doInlining which was not used.