Fix WebGPU ConvTranspose shader bugs for non-vectorizable input channels
Fix two bugs in the pack_input_as4 code path when a_components == 1
(triggered when input channels per group is not divisible by 2 or 4):
1. Weight values were read using x_offset (input tensor offset) instead
of w_offset (weight tensor offset), producing incorrect results.
2. The remainder loop added raw input values to dotProd without
multiplying by the corresponding weight values.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>