[CPU] fix: correct the isa selection by env MAX_CPU_ISA

[azhai219]

Details:

• correct the right isa selection when setting MAX_CPU_ISA
• ...

Tickets:

• [Bug]: Segmentation fault when running inference with oneDNN forced to SSE4.1 (AVX disabled) – possible bug in jit_uni_reorder.cpp for SSE4.1 path #35445

AI Assistance:

• AI assistance used: no / yes
• If yes, summarize how AI was used and what human validation was performed (build/tests/manual checks).

[EgorDuplensky]

@azhai219 Please follow repositories commit subject naming / tagging convention
[CPU]

[rkazants]

Can you please add test? Also, I mention ticket that is already closed.

[maxnick]

Can you please add test? Also, I mention ticket that is already closed.

We already have dedicated tests:
src/plugins/intel_cpu/tests/functional/shared_tests_instances/single_layer_tests/binary_convolution.cpp

The problem with this specific issues is that the Binary Convolution implementation doesn't support (and has never supported) ISA enforce through the runtime info (this technique is used for some other primitves to test all the available implementations using a single HW set). Therefore, to trigger the issue, the test must be run on a CPU with sse42 max ISA, which is a rare case.
Also please note that sse42 is an outdated ISA and is no longer supported in OpenVINO (see https://docs.openvino.ai/2026/about-openvino/release-notes-openvino.html#discontinued-in-2026)

seems tests exist