compiler: fix min/max on floats by using intrinsics

This fixes two edge cases for min/max:
min/max(+0.0, -0.0) = -0.0, +0.0
min/max(number, NaN) = NaN, NaN

The compare and select method does not work here.
I switched to the llvm.minimum/llvm.maximum intrinsics which match the intended behavior.

The integer min/max were also swapped over to using intrinsics.
The compare and select path is now only used by strings.

testdata/go1.21.go

@@ -1,15 +1,25 @@
 package main
 
+import "math"
+
 func main() {
 	// The new min/max builtins.
+	// With int:
 	ia := 1
 	ib := 5
 	ic := -3
+	println("min/max:", min(ia, ib, ic), max(ia, ib, ic))
+	// With float:
 	fa := 1.0
 	fb := 5.0
 	fc := -3.0
-	println("min/max:", min(ia, ib, ic), max(ia, ib, ic))
 	println("min/max:", min(fa, fb, fc), max(fa, fb, fc))
+	// Float +/- 0.0:
+	pos0 := 0.0
+	neg0 := -pos0
+	println("min/max:", min(pos0, neg0), max(pos0, neg0))
+	// Float NaN:
+	println("min/max:", min(math.NaN(), 12.0), max(math.NaN(), 12.0))
 
 	// The clear builtin, for slices.
 	s := []int{1, 2, 3, 4, 5}

testdata/go1.21.txt

@@ -1,5 +1,7 @@
 min/max: -3 5
 min/max: -3.000000e+000 +5.000000e+000
+min/max: -0.000000e+000 +0.000000e+000
+min/max: NaN NaN
 cleared s[:3]: 0 0 0 4 5
 cleared map:    0
 added to cleared map:    four 1