Fix RNG crash on Windows

[vlad-perevezentsev]

This PR adds a Windows workaround for a crash on Battlemage during test runs.

In oneMKL 2026.0 mkl_rng::mt19937 destructor may crash with STATUS_STACK_BUFFER_OVERRUN during DLL_PROCESS_DETACH when using the Level Zero backend and static storage inside a DLL.

The workaround heap-allocates backend_sycl singleton on Windows to avoid running its destructor at shutdown. Although ~backend_sycl() is empty, it triggers destruction of its members (including mkl_rng::mt19937) which causes the crash.
Since backend_sycl is a process-lifetime singleton, skipping its destruction is safe because OS reclaims all
resources at process exit.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• Have you checked performance impact of proposed changes?
• Have you added documentation for your changes, if necessary?
• Have you added your changes to the changelog?

[github-actions]

View rendered docs @ https://intelpython.github.io/dpnp/pull/2871/index.html

[coveralls]

coverage: 78.298% (+0.003%) from 78.295% — fix_rng_crash_windows into master

[github-actions]

Array API standard conformance tests for dpnp=0.20.0dev6=py313h509198e_19 ran successfully.
Passed: 1356
Failed: 4
Skipped: 16

[antonwolfy]

Thank you @vlad-perevezentsev