Refactor the status error message builder.

[ysiraichi]

This PR refactors BuildStatusErrorMessage function, so it's easier to use in other places besides in the MaybeThrow() implementation. For example, when crashing on a status value.

As a practical example, I used it to better show the error information when _get_graph_hash() crashes:

TORCH_SHOW_CPP_STACKTRACES=0

#################################################################
# Before this PR
F0000 00:00:1754594604.459119   30334 init_python_bindings.cpp:3380] Check failed: xtensors_status.ok() _get_graph_hash(): error retrieving the XLA tensors from the given tensor arguments. This is a bug! Please, open an issue in the PyTorch/XLA GitHub repository: https://github.com/pytorch/xla
Status Error: Expected all tensors in the given list to be XLA tensors. Element at index 0 is not an XLA tensor. Got: CPUFloatType
*** Check failure stack trace: ***
    @     0x7497ef2a1159  absl::lts_20230802::log_internal::LogMessageFatal::~LogMessageFatal()
    @     0x7497e3b6bdb8  torch_xla::(anonymous namespace)::InitXlaModuleBindings()::{lambda()#172}::operator()()
    @     0x7497e4225946  pybind11::cpp_function::initialize<>()::{lambda()#3}::_FUN()
    @     0x7497e4204f51  pybind11::cpp_function::dispatcher()
    @     0x74998357aa2e  cfunction_call
Aborted (core dumped)

#################################################################
# After this PR
F0000 00:00:1754593291.250543   19765 init_python_bindings.cpp:3380] Check failed: xtensors_status.ok()

Internal Error:
    _get_graph_hash(): error retrieving the XLA tensors from the given tensor arguments. This is a bug! Please, open an issue in the PyTorch/XLA GitHub repository: https://github.com/pytorch/xla

Status Error:
    Expected all tensors in the given list to be XLA tensors. Element at index 0 is not an XLA tensor. Got: CPUFloatType

*** Check failure stack trace: ***
    @     0x7d80ecea7059  absl::lts_20230802::log_internal::LogMessageFatal::~LogMessageFatal()
    @     0x7d80e176bd76  torch_xla::(anonymous namespace)::InitXlaModuleBindings()::{lambda()#172}::operator()()
    @     0x7d80e1e25276  pybind11::cpp_function::initialize<>()::{lambda()#3}::_FUN()
    @     0x7d80e1e04891  pybind11::cpp_function::dispatcher()
    @     0x7d82acf59a2e  cfunction_call
Aborted (core dumped)

TORCH_SHOW_CPP_STACKTRACES=1

#################################################################
# Before this PR
F0000 00:00:1754594644.896688   30382 init_python_bindings.cpp:3380] Check failed: xtensors_status.ok() _get_graph_hash(): error retrieving the XLA tensors from the given tensor arguments. This is a bug! Please, open an issue in the PyTorch/XLA GitHub repository: https://github.com/pytorch/xla
Status Error: Expected all tensors in the given list to be XLA tensors. Element at index 0 is not an XLA tensor. Got: CPUFloatType

Status Propagation Trace:
    From: GetXlaTensorImpl at torch_xla/csrc/aten_xla_bridge.cpp:88 (error: Failed retrieving the inner XLATensorImpl* from CPUFloatType. It's likely that `tensor` is not an actual XLA tensor, i.e. it wasn't created inside PyTorch/XLA.)
    From: GetXlaTensor at torch_xla/csrc/aten_xla_bridge.cpp:106 (error: Expected XLA tensor. Got: CPUFloatType)
    From: GetXlaTensors at torch_xla/csrc/aten_xla_bridge.cpp:118 (error: Expected all tensors in the given list to be XLA tensors. Element at index 0 is not an XLA tensor. Got: CPUFloatType)

*** Check failure stack trace: ***
    @     0x788d52ca1159  absl::lts_20230802::log_internal::LogMessageFatal::~LogMessageFatal()
    @     0x788d4756bdb8  torch_xla::(anonymous namespace)::InitXlaModuleBindings()::{lambda()#172}::operator()()
    @     0x788d47c25946  pybind11::cpp_function::initialize<>()::{lambda()#3}::_FUN()
    @     0x788d47c04f51  pybind11::cpp_function::dispatcher()
    @     0x788ee6f96a2e  cfunction_call
Aborted (core dumped)

#################################################################
# After this PR
F0000 00:00:1754593329.264671   19813 init_python_bindings.cpp:3380] Check failed: xtensors_status.ok()

Internal Error:
    _get_graph_hash(): error retrieving the XLA tensors from the given tensor arguments. This is a bug! Please, open an issue in the PyTorch/XLA GitHub repository: https://github.com/pytorch/xla

Status Error:
    Expected all tensors in the given list to be XLA tensors. Element at index 0 is not an XLA tensor. Got: CPUFloatType

Status Propagation Trace:
    From: GetXlaTensorImpl at torch_xla/csrc/aten_xla_bridge.cpp:88 (error: Failed retrieving the inner XLATensorImpl* from CPUFloatType. It's likely that `tensor` is not an actual XLA tensor, i.e. it wasn't created inside PyTorch/XLA.)
    From: GetXlaTensor at torch_xla/csrc/aten_xla_bridge.cpp:106 (error: Expected XLA tensor. Got: CPUFloatType)
    From: GetXlaTensors at torch_xla/csrc/aten_xla_bridge.cpp:118 (error: Expected all tensors in the given list to be XLA tensors. Element at index 0 is not an XLA tensor. Got: CPUFloatType)

*** Check failure stack trace: ***
    @     0x73a6bcea7059  absl::lts_20230802::log_internal::LogMessageFatal::~LogMessageFatal()
    @     0x73a6b176bd76  torch_xla::(anonymous namespace)::InitXlaModuleBindings()::{lambda()#172}::operator()()
    @     0x73a6b1e25276  pybind11::cpp_function::initialize<>()::{lambda()#3}::_FUN()
    @     0x73a6b1e04891  pybind11::cpp_function::dispatcher()
    @     0x73a86de37a2e  cfunction_call
Aborted (core dumped)

[zhanyong-wan]

Thanks!