[CodeGen][NPM] MachineScheduler: Inline constructors

[optimisan]

Attempt to fix the compile-time regression introduced by #125703

This seems to make it better on my machine, but I cannot say for sure as the times keep changing (even taking an average over multiple runs)

[optimisan]

@nikic could you check with the tracker bot?

The current procedure to port passes to the NPM is to extract the logic out to an implementation class and create legacy and new (PM) passes that wrap over the implementation class. This introduces the Impl object construction and method call overheads for every pass.

Another approach would be to keep the legacy pass as-is and wrap the new pass over the legacy pass, which would keep the legacy path relatively free of the Impl reference.

[nikic]

It doesn't make any difference: https://llvm-compile-time-tracker.com/compare.php?from=2f7d3ec02349d4503e1fa8c54665025eb4f8e173&to=830818cb44a1e239cfc40035b4a5963c3bd41b43&stat=instructions:u

You can use valgrind to profile locally in order to determine the root cause. Using z51.c from llvm-test-suite may be a good test case.

[optimisan]

Cool, thanks.
And I'll revert that.