fix: handle indices correctly in cshift intrinsic

[HarshitaKalani]

towards: #6370
Also addresses: #6352 (comment)

[HarshitaKalani]

With this, the error is now different:

$ cat lf.sh
lfortran a.f90 --dump-all-passes-fortran --apply-fortran-mangling
gfortran -fcheck=all -g -c pass_fortran_31_insert_deallocate.f90
gfortran pass_fortran_31_insert_deallocate.o -o pass_fortran_31_insert_deallocate
./pass_fortran_31_insert_deallocate
$ sh lf.sh 
At line 10 of file pass_fortran_31_insert_deallocate.f90
Fortran runtime error: Attempt to DEALLOCATE unallocated 'r_g'

Error termination. Backtrace:
#0  0x7a2747823960 in ???
#1  0x7a27478244d9 in ???
#2  0x7a2747824ad6 in ???
#3  0x5953aa69777e in pot3d
        at /home/harshita/Desktop/lfortran/pass_fortran_31_insert_deallocate.f90:10
#4  0x5953aa697a2a in main
        at /home/harshita/Desktop/lfortran/pass_fortran_31_insert_deallocate.f90:10

which is because there are two consecutive deallocate statements after the insert_deallocate pass

! Fortran code after applying the pass: insert_deallocate
program pot3d
implicit none
integer(4), parameter :: dp = 8
integer(4), save :: n = 3
real(8), dimension(:), allocatable :: r_g
allocate(r_g(n))
call genmesh(n, r_g)

! here
deallocate(r_g)
deallocate(r_g) ! Implicit deallocate

contains

...

[gxyd]

The change looks good to me. I'm still thinking if there is a way to add a test case here.

Also, I checked it locally on my machine and this change allows us to remove the workaround from POT3D