[ROCm] Fix GPU memory fault in batched_unary_embeddings backward

[avbokovoy]

Problem

batched_unary_embeddings_test.py::test_gpu crashed with a GPU memory access fault on ROCm. The permute sub-test passed
offsets of length 8 (7 segments) for a T=2 module, violating the op's offsets.numel() == T*B+1 contract. Forward silently truncated
B=3 and dropped the trailing segment; backward walked all 7 segments, deriving an out-of-range table id t=2 that indexed
grad_weight out of bounds.

Fix

• Correct the test offsets to [0..8] (9 entries -> T*B+1 for B=4).
• Add host-side TORCH_CHECK guards on the offsets length in forward (CPU/CUDA/HIP) and backward (CUDA/HIP), so malformed input fails loudly instead of faulting/corrupting grads.

[meta-codesync]

@q10 has imported this pull request. If you are a Meta employee, you can view this in D118492251.

[sylvesterkaczmarek]

The eager paths now reject malformed offsets, but batched_unary_embeddings_forward_meta still floor-divides (offsets.sym_numel() - 1) / T without enforcing the new T > 0 / divisibility invariant. FakeTensor/export can therefore accept an input shape that eager now rejects and report a misleading B. Could the Meta implementation mirror the new validation?

[pytorch-bot]

Workflows were awaiting approval. CI has now been triggered for the ciflow labels on this PR.

[avbokovoy]

The eager paths now reject malformed offsets, but batched_unary_embeddings_forward_meta still floor-divides (offsets.sym_numel() - 1) / T without enforcing the new T > 0 / divisibility invariant. FakeTensor/export can therefore accept an input shape that eager now rejects and report a misleading B. Could the Meta implementation mirror the new validation?

Sure, done in df219f4. Added dedicated test in 99737ef