py/formatfloat: Fix number of digits and exponent sign when rounding.

This patch fixes 2 things when printing a floating-point number that
requires rounding up of the mantissa:
- retain the correct precision; eg 0.99 becomes 1.0, not 1.00
- if the exponent goes from -1 to 0 then render it as +0, not -0

py/formatfloat.c

@@ -376,11 +376,16 @@ int mp_format_float(FPTYPE f, char *buf, size_t buf_size, char fmt, int prec, ch
                 rs[1] = '0';
                 if (e_sign == '-') {
                     e--;
+                    if (e == 0) {
+                        e_sign = '+';
+                    }
                 } else {
                     e++; 
                 }
-            }
+            } else {
+                // Need at extra digit at the end to make room for the leading '1'
                 s++;
+            }
             char *ss = s; 
             while (ss > rs) {
                 *ss = ss[-1];

tests/float/string_format_modulo.py

@@ -44,3 +44,6 @@ print(('%.40g' % 1e-3)[:2])
 print(('%.40g' % 1e-4)[:2])
 
 print("%.0g" % 1) # 0 precision 'g'
+
+print('%.1e' % 9.99) # round up with positive exponent
+print('%.1e' % 0.999) # round up with negative exponent

tests/float/string_format_modulo3.py

@@ -1,4 +1,3 @@
 # uPy and CPython outputs differ for the following
 print("%.1g" % -9.9) # round up 'g' with '-' sign
-print("%.1e" % 9.99) # round up with positive exponent
-print("%.1e" % 0.999) # round up with negative exponent
+print("%.2g" % 99.9) # round up

tests/float/string_format_modulo3.py.exp

@@ -1,3 +1,2 @@
 -10
-1.00e+01
-1.00e-00
+100