trace: improve error handling and error messages.

[ysiraichi]

This PR refactors the trace operation implementation by improving its error message, and returning a status type value.

Key Changes:

• Make tensor_methods::trace return StatusOr<XLATensorPtr>
• Improve error messages and error handling
  • Renamed CheckMMInputIsMatrix to CheckInputIsMatrix
  • Added a new parameter for specifying the operation name, so as to build a better error message

Example

a = torch.rand(2, 2, 2, device="xla")
torch.trace(a)

Before:

Traceback (most recent call last):
  File "examples/trace.py", line 5, in <module>
    torch.trace(a)
RuntimeError: Check failed: (*input_shape_ref).dimensions_size() == 2 (3 vs. 2)invalid argument for trace: expected a matrix (at torch_xla/csrc/tensor_methods.cpp:3579)

Exception raised from operator& at torch_xla/csrc/runtime/tf_logging.cpp:26 (most recent call first):

After:

Traceback (most recent call last):
  File "examples/trace.py", line 5, in <module>
    torch.trace(a)
RuntimeError: trace(): expected the input tensor f32[2,2,2] to be a matrix (i.e. a 2D tensor).

Status Propagation Trace:
    From: CheckInputIsMatrix at torch_xla/csrc/tensor_methods.cpp:491 (error: trace(): expected the input tensor f32[2,2,2] to be a matrix (i.e. a 2D tensor).)
    From: trace at torch_xla/csrc/tensor_methods.cpp:3596
    From: trace at torch_xla/csrc/aten_xla_type.cpp:3874

Exception raised from ThrowStatusError at torch_xla/csrc/status.cpp:128 (most recent call first):