Don't pass -fno-inline-functions when -Oz/-Os are used

[sbc100]

We have no reason not to let clang use its own heuristics here.

As a followup we can consider completely removing the INLINGING_LIMIT setting since it no longer serves any purpose other than to set this clang flag.

[kripken]

This does seem like the right thing to me. However reading the code size changes my impression is the opposite of yours - seems to be more regressions than wins, and some seem significant (e.g. metadce_cxx_noexcept and random_printf). Perhaps it's worth looking into those a little more - might be system libs we should use -Oz over -Os, or even -Oz -fno-inline-functions?

[sbc100]

Adding back -fno-inline-functions to system library builds to avoid any code size regressions.

[kripken]

Thanks, lgtm.

If one of us has time eventually it might be interesting to look into the system lib flags and check if they are optimal. But seems low priority to me.