[patch, libgfortran] Fix PR20257 End of record occurs when writing large arrays

[Jerry DeLisle]

>>
>>
> My first attempt at this was to use a negative number for unit.  IIRC 
> there was specific existing code that generates an error for negative 
> unit numbers.  I will explore this a bit and come back with something.
> 
I must have been thinking of something else.  I just tried -1 for the internal 
unit number and it regression tested fine.  So I think this will do it.

For consistency, I think I will revise is_internal_unit to use the unit number 
instead of a special bit.  If this all tests OK I will commit to trunk and wait 
a while for 4.1 branch.

Let me know if anyone has any other issues.

Regards,

Jerry