Implement faster* BE ABI

[SoniEx2]

We wanted to make wasm3 faster for compute-heavy workloads on big-endian platforms. We can do this by changing the ABI.

Notes: wasi implementation has not been updated for this ABI (wasi tests fail completely), CI has not been updated to run any tests for this ABI but spec tests pass locally.

[vshymanskyy]

I guess this borrows the idea from Wasm2C, which stores the whole linear memory in reverse order?

[SoniEx2]

yeah, that was the idea... maybe it'd be too much of a maintenance burden tho, sorry. >.<

[vshymanskyy]

The idea is great. However, to merge it, we must make it compatible (syntax wise) with the existing imported C functions syntax. I'm not sure if this is possible...

[vshymanskyy]

Oh, looks like it doesn't need any API changes on the Little Endian systems.
It just needs some adjustments of the WASI functions that peek directly into the memory (i.e. we need to reverse strings!).
Am I missing something?

[SoniEx2]

nah, that's pretty much all there is to it, but that's where we ran into #457 and gave up, the changes seemed really big and unapproachable >.<