Add more functorch shards to PR CI (#81919) (#81919)

Summary:
This PR adds functorch shards to some more linux configurations on Pull
Requests. What's missing so far (and coming in the near future) is:
- adding a shard for windows
- adding a shard for asan (functorch currently times out under asan)
- adding shards for things that run in trunk, like mac-os.

X-link: pytorch/pytorch#81919
Approved by: https://github.com/kit1980

Reviewed By: jeanschmidt

Differential Revision: D38067199

Pulled By: zou3519

fbshipit-source-id: eb3356daf003c7fbc82ae06e989373849d69708d

Author: zou3519

test/test_ops.py

@@ -1293,9 +1293,11 @@ def fn(input, weight, bias):
 
         # numerical inconsistencies, look like bugs
         skip('ldexp', dtypes=(torch.float32,), device_type='cpu'),  # fails on all but mac
-        skip('__rmatmul__', dtypes=(torch.float32,), device_type='cpu'),  # fails on all but windows
-        skip('matmul', dtypes=(torch.float32,), device_type='cpu'),  # fails on all but windows
-        skip('nn.functional.conv_transpose3d', dtypes=(torch.float32,)),  # only fails on cpu only linux
+        skip('__rmatmul__', dtypes=(torch.float32,)),  # fails on all but windows
+        skip('matmul', dtypes=(torch.float32,)),  # fails on all but windows
+        skip('nn.functional.conv_transpose3d', dtypes=(torch.float32,)),
+        skip('nn.functional.conv_transpose2d', dtypes=(torch.float32,)),
+        skip('nn.functional.conv_transpose1d', dtypes=(torch.float32,)),
         skip('nn.functional.layer_norm', dtypes=(torch.float32,), device_type='cpu'),  # fails on windows
         skip('linalg.lu_factor', dtypes=(torch.float32,), device_type='cuda'),  # fails on all but windows
         skip('linalg.lu_factor_ex', dtypes=(torch.float32,), device_type='cuda'),  # fails on all but windows