<charconv>: Can general shortest always use Ryu?

When to_chars() is formatting in general shortest mode, it runs Ryu and then switches between fixed and scientific according to the following criterion:

https://github.com/microsoft/STL/blob/8f4c8163775d665d80642044ce27c4bc696127ce/stl/inc/xcharconv_ryu.h#L1333-L1344

When scientific is selected, we directly print Ryu's result. When fixed is selected, we run additional code:

https://github.com/microsoft/STL/blob/8f4c8163775d665d80642044ce27c4bc696127ce/stl/inc/xcharconv_ryu.h#L1389-L1440

When _Can_use_ryu is true, we can use Ryu's result, with a suitably-placed decimal point and zero digits. When _Can_use_ryu is false, we're dealing with a large integer where Ryu rounds away some digits but we need to print all of them exactly, so we activate a fallback algorithm (long division for float, Ryu Printf for double).

When fixed shortest is originally requested, the fallback algorithm is absolutely necessary for some inputs. (The _Max_shifted_mantissa logic is an attempted performance optimization - it precisely detects cases where we can avoid the fallback algorithm. It would be equally correct, but perhaps slower or faster, to activate the fallback algorithm whenever if (_Ryu_exponent > 0) { // case "172900" is true.)

However, when general shortest is originally requested, that's different from unconditionally requesting fixed. I believe that the criterion _Scientific_exponent < 6 implies that we'll never need the fallback algorithm. We need to verify this and adjust the code accordingly.

(Note: plain shortest needs the fallback algorithm because its criterion inspects the Ryu exponent instead of the scientific exponent. For example, the exactly-representable integer 12345678901234567168 is 1.2345678901234567e+19 in scientific shortest. Fixed notation is fewer characters, so plain shortest must use the fallback algorithm for that value.)