Implement 64 bit memory support in the interpreter

[zherczeg]

Large patch (4200 lines) which implements the Memory 64 variant for all memory / load store opcodes in the interpreter. This is not a full support, the memory.grow, memory.init etc. are not implemented yet.

The patch contains a lot of code duplications. Also #394 should land first.

There was no test for simd / atomic + memidx opcode variants, and this patch adds them. There were typos in the code for these variants.

[zherczeg]

The patch is working now. Unfortunately the stack limit needs to be increased to 7MB, because the interpreter stack consumption for debug builds increased a lot. I suspect the new code paths introduced several local variables, and their space is not reused in debug mode.

The reason for the code duplications: memory64 uses 64 bit offsets instead of 32 bit ones, which increases the byte code size, so the original byte codes cannot be reused.

[ksh8281]

Can we allocate stack through bdwgc(we need to support gc, if needs)
when allocate large stack memory for function?

[zherczeg]

Unfortunately, this is not the WebAssembly function. This is the normal C++ interpret function. And only in debug mode. Perhaps I could try to change the macros to reduce the space.

[zherczeg]

On my machine the debug stack space of the interpreter (x86-64) is increased from 0x5bb8 to 0x8098. This is a compiler decision. Perhaps we could try to use less local variables.

[zherczeg]

I have checked the release mode. There is a small stack increase there as well: from 0x128 to 0x1a8 bytes. This amount is negligible compared to the debug stack consumption.

[ksh8281]

I see