broadcast_matmul to handle stride-0 broadcast dimensions

While implementing CPU Grouped-Query Attention, I encountered a limitation with `broadcast_matmul` and stride-0 dimensions. I'm taking a different approach (fused SIMD kernels) for my use case, but wanted to log this for others who may benefit from the fix.

Currently `broadcast_as` creates stride-0 view and `matmul` rejects stride-0 as "non-contiguous". So, I needed to use the workaround: `.broadcast_as(...).contiguous()` but this physically expands memory.

```rust
// GQA: Q has 16 heads, K/V have 8 heads (2 groups)
let q = ...; // [1, 8, 2, 2, 128]
let k = ...; // [1, 8, 1, 128, 2] ← size-1 dim should broadcast

let scores = q.broadcast_matmul(&k)?; // Error: "non-contiguous lhs"
```

I expected it to broadcast dim 2 of rhs (1 → 2) during matmul
But instead I got an error after internal broadcast creates stride-0

Start by tracing the broadcast_matmul, broadcast_as, and matmul entry points to see where stride-0 dimensions become rejected as non-contiguous. Reproduce the shown grouped-query attention shape, then add coverage for broadcasting a size-1 rhs dimension and verify that broadcast_matmul works without requiring contiguous memory expansion.

Written by the indexing model from the issue text.