Update on "[ET-VK] Removed shared memory usage and simplied conv2d dw op shader to improve performance."

This diff removes shared memory usage in `conv2d_dw_output_tile.glsl` shader to improve performance.
Makes sum a one dimensional array, and moves bias application before storing texel.

Differential Revision: [D75499165](https://our.internmc.facebook.com/intern/diff/D75499165/)

[ghstack-poisoned]

Author: trivedivivek

backends/vulkan/vulkan_preprocess.py

@@ -150,14 +150,14 @@ def preprocess(  # noqa: C901
         program = apply_passes(
             program,
             [
+                RemoveRedundantOpsTransform(),
                 AddmmToLinearTransform(),
                 FuseQuantizedOpsTransform(program),
                 SqueezeUnsqueezeInputs(),
                 FuseViewCopyTransform(),
                 ViewCopyToSqueezeUnsqueezePass(),
                 FuseBatchNormWithConvPass(program),
                 FuseClampPass(),
-                RemoveRedundantOpsTransform(),
             ],
         )