Handle meta tensors in FX quantization (pytorch#142262)

Summary:
X-link: pytorch/torchrec#2622


If module being quantized contains a some meta tensors and some tensors with actual device, we should not fail quantization.

Quantization should also not fail if new quantized module is created on a meta device.

If devices contain meta, copying from meta to meta is not necessary, copying from another device to meta can be skipped.

Test Plan:
```
buck run fbcode//mode/dev-nosan fbcode//torchrec/fb/quant/tests:test_embedding_modules
```

Reviewed By: emlin

Differential Revision: D66895899

Author: kausv

torch/ao/quantization/quantize.py

@@ -300,8 +300,8 @@ def insert_activation_post_process(m, special_act_post_process=None):
 
 
 def _get_unique_devices_(module):
-    return {p.device for p in module.parameters()} | {
-        p.device for p in module.buffers()
+    return {p.device for p in module.parameters() if p.device.type != "meta"} | {
+        p.device for p in module.buffers() if p.device.type != "meta"
     }
 
 
@@ -779,7 +779,7 @@ def swap_module(
             # respect device affinity when swapping modules
             devices = _get_unique_devices_(mod)
             assert (
-                len(devices) <= 1
+                len(devices) <= 1 or (len(devices) == 2 and torch.device("meta") in devices)
             ), f"swap_module only works with cpu or single-device CUDA modules, but got devices {devices}"
             device = next(iter(devices)) if len(devices) > 0 else None
             if device: