reduce division to minimum precision necessary

src/ops.rs

@@ -63,7 +63,11 @@ impl std::ops::Div for F8 {
 		}
 		let (m1, e1) = self.split_unbias();
 		let (m2, e2) = rhs.split_unbias();
-		Self::merge_unbias(((m1 as u32) << 24) / m2 as u32, e1 as i32 - e2 as i32 - 24)
+		const BIAS: i32 = M_BITS as i32 + 1;
+		Self::merge_unbias(
+			((m1 as u32) << BIAS) / m2 as u32,
+			e1 as i32 - e2 as i32 - BIAS,
+		)
 	}
 }