[gfortran,patch] Another (possibly last?) LBOUND and UBOUND fix

[FX Coudert]

ping. This fixes two wrong-code PR, and it's not even too hard to  
review :)

> Attached patch fixes both the simplification and translation for  
> intrinsic UBOUND/LBOUND to get correct results in all cases I could  
> imagine. This clears completely PR 29489 & 29712. The logic behind  
> both patches is straightforward, and the only thing noteworthy is  
> that {L,U}BOUND simplification provided by this patch is a bit  
> pessimistic: there are a few more cases that we should be able to  
> simplify here, but for thoses cases the folding of code emitted  
> will take sufficient to not need runtime evaluation.
>
> [...]
>
> OK for mainline, 4.2 and 4.1?