[FEATURE REQUEST] `geru` test coverage and fixtures

[LoserCheems]

Problem statement

The geru kernel currently has no dedicated tests under tests/. There is no test_geru.py to validate that all backends correctly implement the rank-1 update $A = A + \alpha x y^\top$.

Without geru tests:

• regressions or subtle bugs in any backend may go unnoticed,
• CI cannot guarantee numerical equivalence with the Python reference,
• the README BLAS table cannot mark the Test column for geru as complete.

Proposed solution

Add a dedicated pytest module for geru that validates all available backends against the Python reference implementation.

Concretely:

• Introduce tests/test_geru.py alongside other kernel tests.
• For each backend (Python, PyTorch, Triton, CuTe as available), compare outputs with the Python geru reference across various matrix sizes and vector shapes.
• Cover edge cases such as zero scalars, small matrices, and large magnitude values.

Alternatives considered

Manual or informal testing is not sufficient because it:

• is easy to skip,
• does not integrate with CI,
• fails to provide consistent regression coverage.

Implementation details

• Add tests/test_geru.py using pytest conventions.
• Reuse any shared helpers for random input generation and comparison.
• Parameterize tests over matrix dimensions, dtypes, and scalar values.
• Gracefully skip backend-specific tests when backends are unavailable or not yet implemented.

Use case

The geru tests will:

• enforce correctness for geru across all backends,
• support contributors implementing or optimizing rank-1 update kernels,
• align with the project’s emphasis on numerical equivalence.

Related work

• Existing tests: tests/test_copy.py, tests/test_swap.py.
• BLAS geru test suites in other libraries.

Additional context

This issue is part of completing the geru row in the README BLAS table, especially the Test column.