Precision loss for 64-bit Wasm integers

[abelstuker]

Currently, the framework uses JavaScript Number to represent numbers.
This Number type uses the IEEE 754 double precision 64 bit floating point format.
Therefore, this representation results in a loss of precision for 64 bit integers.
More concretely, problems arise when the framework handles i64 (Wasm) values the fall outside the [Number.MIN_SAFE_INTEGER, Number.MAX_SAFE_INTEGER] interval.

The following example demonstrates the problem.
The i64_get_max function returns the largest possible 64-bit integer value.
The i64_eq_max function checks whether the provided argument equals the largest possible 64-bit signed integer value (9223372036854775807).
The i64_eq_max_wrong function checks whether the provided argument equals 9223372036854776000 (the JS Number representation for 9223372036854775807).

(func (export "i64_get_max")        (result i64)  (i64.const 9223372036854775807))
(func (export "i64_eq_max")         (param i64)   (result i32) (i64.eq (local.get 0) (i64.const 9223372036854775807)))
(func (export "i64_eq_max_wrong")   (param i64)   (result i32) (i64.eq (local.get 0) (i64.const 9223372036854776000)))

When executing latch tests, we obtain:

A    FAIL  i64_get_max      | args: []                    | assert_eq: 9223372036854775807n
        failure: Expected 9223372036854775807 got 9223372036854776000

B    PASS  i64_get_max      | args: []                    | assert_eq: 9223372036854775807

C    PASS  i64_get_max      | args: []                    | assert_eq: 9223372036854775500

D    FAIL  i64_eq_max       | args: [9223372036854775807] | assert_eq: 1
        failure: Expected 1 got 0

These results demonstrate the imprecise integer representations of i64 values once present in the framework.
A: The expected return value is correctly represented (as BigInt), but the obtained return value is incorrectly represented.
B: The expected return value (not a BigInt) is incorrectly represented. Also the obtained return value is incorrectly represented (as in A). Hence, the test succeeds (although with incorrect representations).
C: This test shows that mutiple large integers can be rounded (i.e. incorrectly represented) as a single representable JS Number. It shows a second example of a value that equals the obtained result.
D: This test lets the Wasm function check whether the supplied argument equals the largest possible 64-bit signed integer. This check fails.