Update on "[xnnpack] Improve qd8-f16 testing"

* Added a few qd8 fp16 tests for qc4w and qc8w
* Cleaned up and split fp16 and fp32 for basic tests to avoid getting confused regarding atol issues
* Created a TODO to look into atol values

Differential Revision: [D68350952](https://our.internmc.facebook.com/intern/diff/D68350952/)

[ghstack-poisoned]

Author: digantdesai

backends/xnnpack/test/ops/test_linear.py

@@ -584,6 +584,7 @@ def _test_qd8_per_token_weight_per_channel_group_int4(
         for use_bias in [True, False]:
             for M, K, bl, N in zip(M_sizes, K_sizes, bl_sizes, N_sizes):
                 lin_mod = BaseLinear(
+                    in_size=M,
                     input_channels=K,
                     output_channels=N,
                     dtype=dtype,