Fix accuracy issue from fusion (#31404)
### Description of the issue
- symptom:
1. Execution failure on dGPU (onednn) : `[ERROR] [GPU] get_dims() is
called for dynamic shape`
2. Accuracy issue : output video file has garbage out
- root-cause: From Conv-Eltwise-Reshape layers, Eltwise is fused to Conv
and Conv's output format is selected to blocked format 'b_fs_zyx_fsv16'.
But Reshape output format is bfyx with the input blocked format.
<img width="619" height="1019" alt="image"
src="https://github.com/user-attachments/assets/eafa6e39-de5c-44a7-9b26-2b274d0fced0"
/>
Created Reorder to replace existing Reshape which has a mis-match issue
between format and shape
<img width="1157" height="225" alt="image"
src="https://github.com/user-attachments/assets/bd413faf-aeed-4d62-96b9-1fece4d77a98"
/>
- how resolved:
Modified checking logic which used get_dims for dynamic shape in Reduce
layer
Fixed Reshape have simple format for preferred input format.
Fixed a way to add reorder in handle_reshape pass
`I have found Vladimir's comment from existing unit-tests :
// converts tensor to default format with rank = reshape_out_rank
// Likely in the future we'll update that reorder so it will use
reshape_input_rank
// After that expected in format will be bfzyx`
#### The code and line that caused this issue
- Execution failure : reduce_inst.h
- Accuracy issue : handle_reshape.cpp layout_optimizer.cpp
#### Reproduction step and snapshot
- Execute openvino notebooks/wan2.1-text-to-video
#### 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
"*TEST(handle_reshape, reshape_input_reorder)*""
### Tickets:
- *169601*
---------
Signed-off-by: Min, Byung-il <byungil.min@intel.com>
Signed-off-by: Min, Byungil <byungil.min@intel.com>
