Fix sdpa_micro which caused accuracy issue of long-prompt wwb (#34291)
### Description of the issue
- WWB accuracy test with long prompts showed low similarity on multiple
platforms.
<img width="632" height="266" alt="long-wwb"
src="https://github.com/user-attachments/assets/abd82208-66a3-423d-a225-308ddf1122a6"
/>
#### The code and line that caused this issue
- When WWB is executed by MIXED stagemode on the target platforms, sdpa
config is generated by choose_config_xe2 in sdpa_ge_micro.cpp.
This refers to invalid values of 'wg_n_kq' and 'wg_n_vs' which are
pre-defined.
#### Reproduction step and snapshot
- Reproduced by benchmark on dg2-512 or on MTL
`python benchmark.py -t text_gen -d GPU.1 -m
llama-3.2-3b-instruct/pytorch/ov/OV_FP16-INT8_ASYM -pf long-prompt.jsonl
--cb_config "{\"enable_prefix_caching\":true}"`
- cb-config option is required to enable MIXED stage-mode.
- sample long prompt is attached to the ticket
[long-prompt.txt](https://github.com/user-attachments/files/25690220/long-prompt.txt)
#### Checklist
- [x] Is it a proper fix?
- [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? Passed llm_bench
### Tickets:
- CVS-181748, CVS-181751, CVS-181752
### AI Assistance:
- *AI assistance used: no
---------
Signed-off-by: Min, Byungil <byungil.min@intel.com>