Update bounds and docs for the Float trait

Add some bounds to integer types that allow our function trait bounds to
be slightly less verbose. Also clarify documentation and remove a
redundant operation.

src/float/mod.rs

@@ -31,10 +31,10 @@ pub(crate) trait Float:
     + ops::Rem<Output = Self>
 {
     /// A uint of the same width as the float
-    type Int: Int;
+    type Int: Int<OtherSign = Self::SignedInt, UnsignedInt = Self::Int>;
 
     /// A int of the same width as the float
-    type SignedInt: Int;
+    type SignedInt: Int + MinInt<OtherSign = Self::Int, UnsignedInt = Self::Int>;
 
     /// An int capable of containing the exponent bits plus a sign bit. This is signed.
     type ExpInt: Int;
@@ -51,7 +51,7 @@ pub(crate) trait Float:
     /// The bitwidth of the exponent
     const EXPONENT_BITS: u32 = Self::BITS - Self::SIGNIFICAND_BITS - 1;
 
-    /// The maximum value of the exponent
+    /// The saturated value of the exponent (infinite representation), in the rightmost postiion.
     const EXPONENT_MAX: u32 = (1 << Self::EXPONENT_BITS) - 1;
 
     /// The exponent bias value
@@ -83,7 +83,7 @@ pub(crate) trait Float:
     /// Returns true if the sign is negative
     fn is_sign_negative(self) -> bool;
 
-    /// Returns the exponent with bias
+    /// Returns the exponent, not adjusting for bias.
     fn exp(self) -> Self::ExpInt;
 
     /// Returns the significand with no implicit bit (or the "fractional" part)
@@ -175,7 +175,7 @@ macro_rules! float_impl {
             fn normalize(significand: Self::Int) -> (i32, Self::Int) {
                 let shift = significand
                     .leading_zeros()
-                    .wrapping_sub((Self::Int::ONE << Self::SIGNIFICAND_BITS).leading_zeros());
+                    .wrapping_sub(Self::EXPONENT_BITS);
                 (
                     1i32.wrapping_sub(shift as i32),
                     significand << shift as Self::Int,