Update on "Remove unified.py (Quantizer and TwoStepQuantizer ABCs)"

Summary:
Delete `torchao/quantization/unified.py` which defined the `Quantizer`
and `TwoStepQuantizer` abstract base classes. These were trivial ABCs
that only declared method signatures (`quantize`, `prepare`, `convert`)
which all subclasses already implement. Remove the base class
inheritance from all subclasses and clean up imports.

Test Plan:
pytest test/quantization/test_qat.py -x
pytest test/quantization/test_quant_api.py -x

[ghstack-poisoned]

Author: jerryzh168

test/quantization/test_qat.py

@@ -37,6 +37,7 @@
     _replace_linear_8da4w,
     _replace_linear_int4,
 )
+from torchao.quantization.qat import TwoStepQuantizer
 from torchao.quantization.qat.api import (
     ComposableQATQuantizer,
     FromIntXQuantizationAwareTrainingConfig,
@@ -63,7 +64,6 @@
     Int4WeightOnlyQATLinear,
     Int8DynActInt4WeightQATLinear,
 )
-from torchao.quantization.qat.two_step_quantizer import TwoStepQuantizer
 from torchao.quantization.qat.utils import (
     _fake_quantize_per_channel_group,
     _fake_quantize_per_token,