Analyze spectre implications

[twitchyliquid64]

No work has been done to determine if the interpreter/bytecode-dispatch loop is vulnerable to spectre, nor am I confident that mitigations applied for memory load in the native compiler are fully effective.

We should sit down and work out for sure if there is any more work to do here.

My gut feeling is that the dispatch loop has sufficiently high overhead that mis-speculating through incorrect instructions is infeasible. I'm more concerned about I64Load/I32Load in the native compiler, it would be nice to have a test that demonstrates they mitigate spectre.