Fix binary_op int variant (#3745)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/3745

We were using the wrong `sampler3D` meant for `float`/`half`, when we should use `isampler3D` for `int`. [The codegen has a `Dict` mapping](https://www.internalfb.com/code/fbsource/[ac566fa2c8cb64cceb5d47fe4a9c97241a73341d]/fbcode/executorch/backends/vulkan/runtime/api/gen_vulkan_spv.py?lines=58-71) to templatize this logic. Since we're already here, modernize to the new layout generation functions.

## Before
```
layout(set = 0, binding = 1) uniform PRECISION sampler3D image_in;
layout(set = 0, binding = 2) uniform PRECISION sampler3D image_other;
```
Input values were read incorrectly and got wild results.
```
ET-VK
tensor([[-2147483648, -2147483648,           0],
        [-2147483648,           0,           0]], dtype=torch.int32)
Eager-mode
tensor([[-123, -131,  182],
        [ -29,   24,  138]], dtype=torch.int32)
````

## After
```
layout(set = 0, binding = 1) uniform PRECISION ${SAMPLER_T[NDIM][DTYPE]} image_in;
layout(set = 0, binding = 2) uniform PRECISION ${SAMPLER_T[NDIM][DTYPE]} image_other;
```
OR modernize to the new layout generation functions, which already use and abstract `SAMPLER_T` to avoid the same future mistake.
```
${layout_declare_tensor(1, "r", "t_in", DTYPE, STORAGE)}
${layout_declare_tensor(2, "r", "t_other", DTYPE, STORAGE)}
```
ghstack-source-id: 227761995

Reviewed By: nathanaelsee

Differential Revision: D57731470

fbshipit-source-id: f233089e5b4ced6f3d47b90a9c84795cf62bb4de