[GPU] Fix reorder eltwise broadcast (#31221)
### Description of the issue
- symptom: Functional issue on dGPU(onednn) and accuracy issue on
iGPU(cldnn execution).
`Segmentation fault`
`Exact Match accuracy : 77.87% Mean Iou Score: 0.7564`
- root-cause: Target layer 'Add_9569' takes different dimension inputs
with dynamic shape for NUMPY auto-broadcasting.
Currently, no logic for this NUMPY broadcasting implemented.
add_required_reorder pass tried to add a Reorder layer from bfyx to
b_fs_zyx_fsv16 which is not supported.
This caused both execution failure and accuracy issue.
Also found Reorder kernel does not properly handle -1x-1x2 shape size
for 5dims blocked format.
- how resolved: When NUMPY auto-broadcating of Eltwise layer is coming
for blocked format, then Reorder is not added to prevent invalid
reordering of smaller tensor size to the blocked format.
<img width="1872" height="570" alt="image"
src="https://github.com/user-attachments/assets/d5581ea1-0d4d-425d-a8d4-20153a8743ac"
/>
static :
<img width="1251" height="823" alt="image"
src="https://github.com/user-attachments/assets/519d4ae4-fe6c-4a26-91f8-7dda1d2b8557"
/>
Result : `Exact Match Accuracy : 100.00% Mean IoU Score : 0.9979`
#### The code and line that caused this issue
- No target format supported : reorder_impls.cpp
- Added logic into : primitive_inst.cpp reorder_inputs.cpp
#### Reproduction step and snapshot
- Download attached package
[06585866](https://intel-my.sharepoint.com/:f:/p/sahira_rizvi/Eqfmh_PV9EZJnlmGsurqHrgBor5F4cXNvK-ZMkwp9ylFqg?e=AoycQy)
from ticket
`$ python yolov5/ov_detect.py`
`$ python compare.py OV_CPU OV_GPU`
#### Checklist
- [x] Is it a proper fix? (not a workaround)
- [x] Did you include test case for this fix, if necessary?
- [x] Did you review existing test that can be extended to cover this
scenario? Which test did you review? : reviewed unittests with
"*broadcast_test_two_inputs_blocked_format*" and
"align_shape_for_numpy_broadcast_test"
### Tickets:
- *169054*, *171172*
---------
Signed-off-by: Min, Byungil <byungil.min@intel.com>
