Bugzilla

From the Chrome bug:
```Current performance penalty init_stack_vars = true:
Binary size increase: 1.5%
linux-perf/speedometer2 Total Score: 2.5% slower
linux-perf/blink_perf.dom and linux-perf/dromaeo regression vary from 0 to 5%

FYI. Zero initialization:
Binary size increase: 1.05%
linux-perf/speedometer2 Total Score: 0.5% faster
linux-perf/blink_perf.dom and linux-perf/dromaeo from up 12% speedup
                                                 to 4.5% slowdown
```

Also, I see a bunch of followup bugs to fix 5-15% regressions in things like VP9 encode, CSS, etc (disabling the flag for those I think).

The perfherder results need more runs; most are "medium" confidence - small differences need more runs for statistical confidence.  I see small regressions is most speedindex numbers, and other performance numbers, typically in the 1-6% range.

Some focused benchmarks show large regressions, like WebAudio - up to 33%: https://treeherder.mozilla.org/perfherder/comparesubtest?originalProject=try&newProject=try&newRevision=7f330cc417d914f6fd6a0107307884712a5bca7b&originalSignature=3442639&newSignature=3442639&framework=13&originalRevision=ef2d1eaa5a80d405df665726cc3e6497289db79d&page=1

Taking this sort of hit on pageload would be concerning, and would require a lot more looking and thought about the impact (and pointwise mitigations/disabling).