[fx2trt] support ops for hf_T5 (#62)

Summary:
X-link: pytorch/fx2trt#62

1. support type_as()
2. improve power with method
3. improve to_dtype(torch.Tensor)
4. support float(),half(),int()
5. support isinf(). Not possible to support isnan()
6. support torch.any()
7. support rsqrt()

(Note: this ignores all push blocking failures!)

Test Plan: Test passed

Reviewed By: 842974287

Differential Revision: D35516820

fbshipit-source-id: 7f4b7c1b833034121f7643565bd4d1b4133319bf

Author: Wei Wei

torch/testing/_internal/common_fx2trt.py

@@ -208,7 +208,7 @@ def run_test(
 
         if test_implicit_batch_dim:
             interp = TRTInterpreter(mod, InputTensorSpec.from_tensors(inputs))
-            super().run_test(mod, inputs, expected_ops, unexpected_ops, interp, rtol, atol)
+            super().run_test(mod, inputs, expected_ops, unexpected_ops, interp, rtol, atol, precision)
 
         if test_explicit_batch_dim:
             interp = TRTInterpreter(