Use nanf() instead of nan() in single-precision float libm math functions

This patch reduces code size for a few single-precision float math
functions, by using nanf() instead of nan() where required.

newlib/libm/math/wf_acos.c

@@ -31,7 +31,7 @@
 	if(fabsf(x)>1.0f) {
 	    /* acosf(|x|>1) */
 	    errno = EDOM;
-	    return (float) nan("");
+	    return nanf("");
 	} else
 	    return z;
 #endif

newlib/libm/math/wf_asin.c

@@ -38,7 +38,7 @@
 	if(fabsf(x)>1.0f) {
 	    /* asinf(|x|>1) */
 	    errno = EDOM;
-	    return (float)nan("");
+	    return nanf("");
 	} else
 	    return z;
 #endif

newlib/libm/math/wf_log.c

@@ -41,7 +41,7 @@
 	} else { 
 	    /* logf(x<0) */
 	    errno = EDOM;
-	    return nan("");
+	    return nanf("");
         }
 #endif
 }

newlib/libm/math/wf_log10.c

@@ -41,7 +41,7 @@
 	    } else { 
 		/* log10f(x<0) */
 		errno = EDOM;
-		return nan("");
+		return nanf("");
             }
 	} else
 	    return z;