[GPU] Fix Gemma4-E4B SDPA model

[Lyamin-Roman]

Details:

This PR fixes two consecutive errors related to the inference of Gemma4-E4B SDPA version model on iGPU

1. In this model, a single KVCache is shared between multiple SDPA layers
   So problems arise in SyncInferRequest::allocate_states and VariableStateIndirectKVCacheCompressed is created when one kv_cache_prim is compressed and the other is not, which leads to output mismatches in the plugin.
   Currently, this only affects iGPUs with supports_immad=false, so we can temporarily fix it by disabling KVCacheCompression for such a graph, which removes potential mismatches.
   It will need to be redone when PA models are enabled.

2. Out-of-bounds SLM access in sdpa_opt finalization kernel.
   The finalization stage allocated tmp_slm[SUBGROUP_SIZE] elements for cross-subgroup reduction, but the actual number of subgroups per workgroup is SUBGROUPS_PER_WG = CEIL_DIV(V_HEAD_SIZE * SG_SCALE_FACTOR, SUBGROUP_SIZE).
   When V_HEAD_SIZE > SUBGROUP_SIZE^2 (e.g. head_size=512, SUBGROUP_SIZE=16 gives SUBGROUPS_PER_WG=32 > 16), tmp_slm[sgid] writes go out of bounds.
   So changed tmp_slm allocation to SUBGROUPS_PER_WG and replaced the single-pass lane-indexed reduction with a folded loop over CEIL_DIV(SUBGROUPS_PER_WG, SUBGROUP_SIZE) iterations, correctly reducing across all subgroups regardless of head size.

   There are problems reproducing the problem using the test, apparently more interactions with memory are needed to reproduce it as in the inference of the whole model

AI Assistance:

• AI assistance used: yes

[clee30]

As this only affect iGPU with support_immad = false, you may skip the check if support_immad=true

[Lyamin-Roman]

This transformation is enabled only in the following cases:

1. is_paged_attention_model
2. info.supports_immad == false
3. auxiliary_kv_update_model

And when a model with PA appears, I think the same problem will appear, so I think it's better to temporarily disable this transformation overall for the unsupported graph