[PyTorch Backend] [2.67.0/container 26.03] Torch models log 'ModelInitialize' on each request

After updating our deployment container version to 26.03 we noticed lots of new logs (INFO) reporting model initialization (on every request) for models using pytorch backend. ``` I0407 13:11:47.653802 1 libtorch.cc:136] "TRITONBACKEND_ModelInitialize: ... I0407 13:11:47.653828 1 libtorch.cc:210] "TRITONBACKEND_ModelInstanceInitialize: ... I0407 13:11:47.766677 1 libtorch.cc:136] "TRITONBACKEND_ModelInitialize: ... I0407 13:11:47.766705 1 libtorch.cc:210] "TRITONBACKEND_ModelInstanceInitialize: ... I0407 13:11:47.884304 1 libtorch.cc:136] "TRITONBACKEND_ModelInitialize: ... I0407 13:11:47.884331 1 libtorch.cc:210] "TRITONBACKEND_ModelInstanceInitialize: ... ``` Looking at the code, it seems to [originate in code checking for PT2 format support](https://github.com/triton-inference-server/pytorch_backend/blob/8ffd6b20804956bd1fe990d3426e3cd1cc46721a/src/libtorch.cc#L137). So question - does it make sense to log this when it's just checking whether PT2 path should be run ? CC author: @whoisj