[BUG]Error log is misleading when IPU device is inaccessible

[tianboh]

Describe the bug
As title.

Screenshots

2022-01-06T06:45:13.321528Z popart:popart 904.904 I: Start to create Executable...
2022-01-06T06:45:13.321620Z popart:popart 904.904 I: Computation is not initialized. init it first
2022-01-06T06:45:13.321629Z popart:popart 904.904 I: PIPELINE not used for this run, Set the global virtual group to ipu 0
2022-01-06T06:45:18.424014Z popart:popart 904.904 E: Must pass a valid deviceInfo to InferenceSession::createFromOnnxModel

[0] _odla_computation::init(bool)
[1] odla_CreateExecutable
[2] tensorflow::halo::MlirIpuExecutable::Init()
[3] tensorflow::halo::HaloCompilationCache::CompileImpl(std::unique_ptr<tensorflow::halo::HaloCompilerInput, std::default_delete<tensorflow::halo::HaloCompilerInput> >, tensorflow::NameAttrList const&, std::map<int, tensorflow::Tensor, std::less<int>, std::allocator<std::pair<int const, tensorflow::Tensor> > > const&, std::set<int, std::less<int>, std::allocator<int> > const&, std::map<int, tensorflow::halo::OptionalTensor, std::less<int>, std::allocator<std::pair<int const, tensorflow::halo::OptionalTensor> > > const&, tensorflow::OpKernelContext*, tensorflow::halo::Executable**)
[4] tensorflow::halo::HaloCompilationCache::Compile(std::unique_ptr<tensorflow::halo::HaloCompilerInput, std::default_delete<tensorflow::halo::HaloCompilerInput> >, tensorflow::NameAttrList const&, std::map<int, tensorflow::Tensor, std::less<int>, std::allocator<std::pair<int const, tensorflow::Tensor> > > const&, std::set<int, std::less<int>, std::allocator<int> > const&, std::map<int, tensorflow::halo::OptionalTensor, std::less<int>, std::allocator<std::pair<int const, tensorflow::halo::OptionalTensor> > > const&, tensorflow::OpKernelContext*, tensorflow::halo::Executable**)
[5] tensorflow::halo::HaloLaunchOp::CompileToLocalExecutable(tensorflow::OpKernelContext*, tensorflow::NameAttrList const&, std::map<int, tensorflow::halo::OptionalTensor, std::less<int>, std::allocator<std::pair<int const, tensorflow::halo::OptionalTensor> > >*, tensorflow::halo::Executable**)
[6] tensorflow::halo::HaloLaunchOp::CompileAndRun(tensorflow::OpKernelContext*, tensorflow::halo::DoneHelper*, bool*)
[7] tensorflow::halo::HaloLaunchOp::ComputeAsync(tensorflow::OpKernelContext*, std::function<void ()>)



2022-01-06T06:45:18.424154Z popart:popart 904.904 E: Session::createFromOnnxModel failed:Must pass a valid deviceInfo to InferenceSession::createFromOnnxModel
2022-01-06T06:45:18.494088Z popart:popart 904.904 E: Failed to init computation when compiling.
2022-01-06 14:45:18.637901: I tensorflow/compiler/bridge/kernels/halo_launch.cc:524] Done Local compile:Internal: Create executable failed
2022-01-06 14:45:18.639146: W tensorflow/compiler/bridge/kernels/halo_launch.cc:436] cluster: cluster_0_halo_launch fallbacks to TF as Create executable failed
2022-01-06 14:45:18.639214: I tensorflow/compiler/bridge/kernels/halo_launch.cc:440] cluster: cluster_0_halo_launch fallback to TF.
2022-01-06 14:45:18.640200: W tensorflow/compiler/bridge/kernels/halo_launch.cc:378] HaloLaunch execute function failed: Handle: 140125997346536 not found.
Encount error(s) when warmup: Handle: 140125997346536 not found.
	 [[{{node cluster_0_halo_launch}}]]