[ExecuTorch][WebGPU] et_vk.sdpa: shape-route QK to a per-entry kernel for channel attention (15-30x faster) #10886

Job	Run time
test-arm / test-backend-linux (arm_vgf_fp, models) / linux-job	1m 39s
test-arm / test-backend-linux (arm_tosa_fp, models) / linux-job	1m 34s
test-arm / test-backend-linux (arm_tosa_fp, operators) / linux-job	1m 41s
test-arm / test-backend-linux (arm_vgf_fp, operators) / linux-job	1m 28s
test-arm / test-backend-macos	-9s
test-arm / package-golden-artifacts	0s
	6m 13s

Provide feedback