val.call<T>() can't return 64-bit integers

[RReverser]

Simple demo (can be reproduced in test_i64_val.cpp):

val::global().call<int64_t>("BigInt", max_int64_t);

fails with:

TypeError: Cannot convert a BigInt value to a number
    at wasm://wasm/000d1786:wasm-function[165]:0x66f9
    at wasm://wasm/000d1786:wasm-function[85]:0x44b2
    at wasm://wasm/000d1786:wasm-function[55]:0x381d
    at wasm://wasm/000d1786:wasm-function[41]:0x2810
    at wasm://wasm/000d1786:wasm-function[180]:0x6ad0
    at D:\emscripten\out\test\test_i64_val.js:629:12
    at callMain (D:\emscripten\out\test\test_i64_val.js:5377:15)
    at doRun (D:\emscripten\out\test\test_i64_val.js:5430:24)
    at run (D:\emscripten\out\test\test_i64_val.js:5443:5)
    at removeRunDependency (D:\emscripten\out\test\test_i64_val.js:584:7)
    at receiveInstance (D:\emscripten\out\test\test_i64_val.js:734:5)
    at receiveInstantiationResult (D:\emscripten\out\test\test_i64_val.js:752:12)
    at createWasm (D:\emscripten\out\test\test_i64_val.js:778:17)
Thrown at:
    at $func165 (wasm://wasm/000d1786:1:26362)
    at $func85 (wasm://wasm/000d1786:1:17587)
    at $func55 (wasm://wasm/000d1786:1:14366)
    at $func41 (wasm://wasm/000d1786:1:10257)
    at $main (wasm://wasm/000d1786:1:27345)
    at D:\emscripten\out\test\test_i64_val.js:629:12
    at callMain (D:\emscripten\out\test\test_i64_val.js:5377:15)
    at doRun (D:\emscripten\out\test\test_i64_val.js:5430:24)
    at run (D:\emscripten\out\test\test_i64_val.js:5443:5)
    at removeRunDependency (D:\emscripten\out\test\test_i64_val.js:584:7)

This is due to generic wiretype being a double. I imagine the easiest fix would be changing so that the return value is stored on the stack instead of being actually returned from the function, so that you can store bigints via HEAP64 and all other numbers via HEAPF64 in the same ABI. cc @brendandahl

[sbc100]

Seems a shame to have to go via linear memory when we have wasm bigint integration enabled by default these days.

[RReverser]

Yeah but the alternative is having separate ABIs for "all types" and "int64" like emval_as.

I actually have a refactoring branch that makes this all a lot simpler and fixes bugs like this one as a "side effect".

Problem is, it increases the generated code size a bit - which is why I never landed that - but maybe I should... Code size golfing is fun but the simplicity might be worth it.