Fix possible memory leaks in backend

[ndgrigorian]

This PR fixes a few possible memory leaks in the mkl_fft backend caused by possible allocation failures, which could leak structs or the DFTI capsule

These were caught during free-threaded Python review, with some surfacing in free-threaded test runs, where a thread could fail to allocate the capsule and then cache the failed allocation, leading to the entire thread becoming poisoned

[Copilot]

🟢 Approval recommended

The changes are tightly scoped, consistently propagate allocation failures, and I did not find any remaining unchecked call sites or new leak paths in the modified regions.

Pull request overview

This PR hardens mkl_fft’s C/Cython backend against allocation-failure paths to avoid leaking iterator/capsule state and to surface memory exhaustion as a proper Python MemoryError, aligning with the PR’s goal of preventing “poisoned” thread-local cache state under free-threaded stress.

Changes:

• Make multi_iter_new / multi_iter_masked_new return an error code and ensure failed iterators are non-iterable and safely destructible.
• Add allocation checks in mklfft.c.src (mask/shape/stride buffers), propagate DFTI_MEMORY_ERROR, and fix an assert that accidentally used assignment.
• Make TLS DFTI cache capsule creation exception-safe and centralize MKL status-to-exception mapping (raising MemoryError for DFTI_MEMORY_ERROR).

File summaries

File	Description
mkl_fft/tests/test_fftnd.py	Switches a few assertion messages to f-strings (no behavior change).
mkl_fft/src/multi_iter.h	Iterator constructors now fail fast on allocation errors and are safe to free after partial initialization.
mkl_fft/src/mklfft.c.src	Adds allocation-failure handling to prevent leaks and correctly propagates memory errors; fixes an incorrect assert.
mkl_fft/_pydfti.pyx	Prevents TLS capsule caching of failed allocations and maps MKL memory errors to Python MemoryError.

Review details

• Files reviewed: 5/5 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[jharlow-intel]

only thing I can really note is the broken markdown github links

[antonwolfy]

Thank you @ndgrigorian, LGTM

Change was made