Update on "[ET-VK] Implement linear_qcs4w"

## Context

Title says it all!

## Changes

Extended the implementation of `linear_qcsnw` to support packed 4-bit weight tensors.

Differential Revision: [D73941991](https://our.internmc.facebook.com/intern/diff/D73941991/)

[ghstack-poisoned]

Author: SS-JIA

backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.yaml

@@ -28,3 +28,11 @@ linear_qcsnw_coop:
       OUT_STORAGE: buffer
       WEIGHT_STORAGE: buffer
       SCALES_STORAGE: buffer
+    - NAME: linear_qcs4w_coop_texture3d_texture3d_texture2d_texture2d_float
+      TILE_TXCOLS: 2
+      QUANT_NBITS: 4
+    - NAME: linear_qcs4w_coop_buffer_buffer_texture2d_texture2d_float
+      IN_STORAGE: buffer
+      OUT_STORAGE: buffer
+      TILE_TXCOLS: 2
+      QUANT_NBITS: 4