Fix precision bug for LONGREAL literal.

Emit enough significant digits for the C compiler to re-parse the exact binary value: a 64-bit LONGREAL needs 17 (DBL_DECIMAL_DIG).  The old 13 truncated the literal, so C parsed a rounded, different double — a C-backend float-literal precision bug.

Author: hosking

m3-sys/m3middle/src/TFloat.m3

@@ -165,7 +165,7 @@ PROCEDURE ToChars (READONLY f: Float;  VAR buf: ARRAY OF CHAR): INTEGER =
     IF f.pre = Precision.Short THEN
       len := Convert.FromFloat (zz, FLOAT(f.fraction, REAL), 13, Convert.Style.Sci);
     ELSIF f.pre = Precision.Long THEN
-      len := Convert.FromLongFloat (zz, FLOAT(f.fraction, LONGREAL), 13, Convert.Style.Sci);
+      len := Convert.FromLongFloat (zz, FLOAT(f.fraction, LONGREAL), 17, Convert.Style.Sci);
     ELSE
       len := Convert.FromExtended (zz, f.fraction, 36, Convert.Style.Sci);
     END;