Floats formatted with "correct" precision

Conversion float->string ensures that, for any float f,
tonumber(tostring(f)) == f, but still avoiding noise like 1.1
converting to "1.1000000000000001".

luaconf.h

@@ -416,8 +416,13 @@
 @@ l_floatatt(x) corrects float attribute 'x' to the proper float type
 ** by prefixing it with one of FLT/DBL/LDBL.
 @@ LUA_NUMBER_FRMLEN is the length modifier for writing floats.
-@@ LUA_NUMBER_FMT is the format for writing floats.
-@@ lua_number2str converts a float to a string.
+@@ LUA_NUMBER_FMT is the format for writing floats with the maximum
+** number of digits that respects tostring(tonumber(numeral)) == numeral.
+** (That would be floor(log10(2^n)), where n is the number of bits in
+** the float mantissa.)
+@@ LUA_NUMBER_FMT_N is the format for writing floats with the minimum
+** number of digits that ensures tonumber(tostring(number)) == number.
+** (That would be LUA_NUMBER_FMT+2.)
 @@ l_mathop allows the addition of an 'l' or 'f' to all math operations.
 @@ l_floor takes the floor of a float.
 @@ lua_str2number converts a decimal numeral to a number.
@@ -428,8 +433,6 @@
 
 #define l_floor(x)		(l_mathop(floor)(x))
 
-#define lua_number2str(s,sz,n)  \
-	l_sprintf((s), sz, LUA_NUMBER_FMT, (LUAI_UACNUMBER)(n))
 
 /*
 @@ lua_numbertointeger converts a float number with an integral value
@@ -458,6 +461,7 @@
 
 #define LUA_NUMBER_FRMLEN	""
 #define LUA_NUMBER_FMT		"%.7g"
+#define LUA_NUMBER_FMT_N	"%.9g"
 
 #define l_mathop(op)		op##f
 
@@ -474,6 +478,7 @@
 
 #define LUA_NUMBER_FRMLEN	"L"
 #define LUA_NUMBER_FMT		"%.19Lg"
+#define LUA_NUMBER_FMT_N	"%.21Lg"
 
 #define l_mathop(op)		op##l
 
@@ -488,7 +493,8 @@
 #define LUAI_UACNUMBER	double
 
 #define LUA_NUMBER_FRMLEN	""
-#define LUA_NUMBER_FMT		"%.14g"
+#define LUA_NUMBER_FMT		"%.15g"
+#define LUA_NUMBER_FMT_N	"%.17g"
 
 #define l_mathop(op)		op