Compare 0-dimension tensor got CUDA_ERROR_INVALID_VALUE

I got error CUDA_ERROR_INVALID_VALUE when running the code on cuda device. But the same code is running fine in cpu device.

```
use candle_core::{Device, DType, Tensor};

fn main() -> Result<(), Box> {
// let device = Device::Cpu;
let device = Device::new_cuda(0)?;

let a = Tensor::zeros((2, 1, 1, 0), DType::F32, &device)?;
let b = Tensor::zeros((2, 1, 1, 0), DType::F32, &device)?;

let diff = a.eq(&b)?;
println!("{diff}");
Ok(())
}
```

Error message:

Error: WithBacktrace { inner: Cuda(Cuda(DriverError(CUDA_ERROR_INVALID_VALUE, "invalid argument"))), backtrace: Backtrace [{ fn: "candle_core::error::Error::bt", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/error.rs", line: 231 }, { fn: " as candle_core::cuda_backend::error::WrapErr>::w::{{closure}}", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/cuda_backend/error.rs", line: 60 }, { fn: "core::result::Result::map_err", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/core/src/result.rs", line: 829 }, { fn: " as candle_core::cuda_backend::error::WrapErr>::w", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/cuda_backend/error.rs", line: 60 }, { fn: "::f", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/cuda_backend/mod.rs", line: 968 }, { fn: "candle_core::cuda_backend::utils::Map2Any::map", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/cuda_backend/utils.rs", line: 166 }, { fn: "::cmp", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/cuda_backend/mod.rs", line: 1261 }, { fn: "candle_core::storage::Storage::cmp", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/storage.rs", line: 142 }, { fn: "candle_core::tensor::Tensor::cmp", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/tensor.rs", line: 991 }, { fn: "candle_core::tensor::Tensor::eq", file: "/root/.cargo/git/checkouts/candle-0c2b4fa9e5801351/a226a97/candle-core/src/tensor.rs", line: 1000 }, { fn: "myapp::main", file: "./src/main.rs", line: 10 }, { fn: "core::ops::function::FnOnce::call_once", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/core/src/ops/function.rs", line: 250 }, { fn: "std::sys_common::backtrace::__rust_begin_short_backtrace", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/sys_common/backtrace.rs", line: 155 }, { fn: "std::rt::lang_start::{{closure}}", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/rt.rs", line: 159 }, { fn: "core::ops::function::impls:: for &F>::call_once", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/core/src/ops/function.rs", line: 284 }, { fn: "std::panicking::try::do_call", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/panicking.rs", line: 559 }, { fn: "std::panicking::try", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/panicking.rs", line: 523 }, { fn: "std::panic::catch_unwind", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/panic.rs", line: 149 }, { fn: "std::rt::lang_start_internal::{{closure}}", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/rt.rs", line: 141 }, { fn: "std::panicking::try::do_call", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/panicking.rs", line: 559 }, { fn: "std::panicking::try", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/panicking.rs", line: 523 }, { fn: "std::panic::catch_unwind", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/panic.rs", line: 149 }, { fn: "std::rt::lang_start_internal", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/rt.rs", line: 141 }, { fn: "std::rt::lang_start", file: "/rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/rt.rs", line: 158 }, { fn: "main" }, { fn: "__libc_start_main" }, { fn: "_start" }] }

Reproduce the CUDA failure with the example in the issue, then inspect Tensor::eq in candle_core/src/tensor.rs and the CUDA comparison path in candle_core/src/cuda_backend/mod.rs, including utils.rs. Confirm that comparing the two zero-sized tensors succeeds on CUDA as it does on CPU, and add or update coverage where the existing comparison behavior is exercised.

Written by the indexing model from the issue text.