Add relaxed float to int conversions (#25)

For #21.

proposals/relaxed-simd/Overview.md

@@ -127,6 +127,42 @@ def relaxed_i8x16_swizzle(a, s):
     return result
 ```
 
+### Float/Double to int conversions
+
+- `relaxed i32x4.trunc_f32x4_s` (relaxed version of `i32x4.trunc_sat_f32x4_s`)
+- `relaxed i32x4.trunc_f32x4_u` (relaxed version of `i32x4.trunc_sat_f32x4_u`)
+- `relaxed i32x4.trunc_f64x2_s_zero` (relaxed version of `i32x4.trunc_sat_f64x2_s_zero`)
+- `relaxed i32x4.trunc_f64x2_u_zero` (relaxed version of `i32x4.trunc_sat_f64x2_u_zero`)
+
+These instructions have the same behavior as the non-relaxed instructions for
+lanes that are in the range of an `i32` (signed or unsigned depending on the
+instruction). For lanes that contain values which are out of bounds or NaN, the
+result is implementation-defined.
+
+```python
+def relaxed_i32x4_trunc_f32x4(a : f32x4, signed : bool) -> i32x4:
+    result = []
+    min = signed ? INT32_MIN : UINT32_MIN
+    max = signed ? INT32_MAX : UINT32_MAX
+    for i in range(4):
+      r = truncate(a[i])
+      if min <= r <= max:
+        result[i] = r
+      else:
+        result[i] = UNDEFINED
+
+def relaxed_i32x4_trunc_f64x2_zero(a : f64x2, signed : bool) -> i32x4:
+    result = [0, 0, 0, 0]
+    min = signed ? INT32_MIN : UINT32_MIN
+    max = signed ? INT32_MAX : UINT32_MAX
+    for i in range(2):
+      r = truncate(a[i])
+      if min <= r <= max:
+        result[i] = r
+      else:
+        result[i] = UNDEFINED
+```
+
 ## Binary format
 
 > This is a placeholder for binary format of instructions and new constructs.