Fix wasmi no_std support (wasmi-labs#218)

Two issues are fixed:

 1. num-bigint 0.2 requires std, and is used to perform float to int conversions
    since wasmi-labs#186. This patch disables this change in no_std environments.
    This change can be reverted once num-bigint 0.3 is released, which will
    support no_std.

 2. The `f{32,64}::fract` method is currently not implemented by core, only std.
    This patch uses the no_std supporting implementation from num-trait's
    FloatCore trait.

Cargo.toml

@@ -15,8 +15,8 @@ wasmi-validation = { version = "0.3", path = "validation", default-features = fa
 parity-wasm = { version = "0.41.0", default-features = false }
 memory_units = "0.3.0"
 libm = { version = "0.1.2", optional = true }
-num-rational = "0.2.2"
-num-traits = "0.2.8"
+num-rational = { version = "0.2.2", default-features = false }
+num-traits = { version = "0.2.8", default-features = false }
 libc = "0.2.58"
 errno = { version = "0.2.4", optional = true }
 
@@ -32,6 +32,7 @@ std = [
     "parity-wasm/std",
     "wasmi-validation/std",
     "num-rational/std",
+    "num-rational/bigint-std",
     "num-traits/std"
 ]
 # Enable for no_std support

src/nan_preserving_float.rs

@@ -2,6 +2,7 @@
 
 use core::cmp::{Ordering, PartialEq, PartialOrd};
 use core::ops::{Add, Div, Mul, Neg, Rem, Sub};
+use num_traits::float::FloatCore;
 
 macro_rules! impl_binop {
     ($for:ident, $is:ident, $op:ident, $func_name:ident) => {
@@ -63,7 +64,7 @@ macro_rules! float {
             }
 
             pub fn fract(self) -> Self {
-                self.to_float().fract().into()
+                FloatCore::fract(self.to_float()).into()
             }
 
             pub fn min(self, other: Self) -> Self {

src/value.rs

@@ -368,13 +368,22 @@ impl WrapInto<F32> for F64 {
 macro_rules! impl_try_truncate_into {
     (@primitive $from: ident, $into: ident, $to_primitive:path) => {
         impl TryTruncateInto<$into, TrapKind> for $from {
+            // FIXME: BigRational uses the num-bigint crate, which requires std
+            // as of version `0.2.6`. `num-bigint` 0.3 will have no_std support.
+            #[cfg(feature = "std")]
             fn try_truncate_into(self) -> Result<$into, TrapKind> {
                 // Casting from a float to an integer will round the float towards zero
                 num_rational::BigRational::from_float(self)
                     .map(|val| val.to_integer())
                     .and_then(|val| $to_primitive(&val))
                     .ok_or(TrapKind::InvalidConversionToInt)
             }
+
+            #[cfg(not(feature = "std"))]
+            fn try_truncate_into(self) -> Result<$into, TrapKind> {
+                // Casting from a float to an integer will round the float towards zero
+                $to_primitive(&self).ok_or(TrapKind::InvalidConversionToInt)
+            }
         }
     };
     (@wrapped $from:ident, $intermediate:ident, $into:ident) => {