PyTorch's FakeTensor adds hinted storage size metadata for tracing

The PyTorch team has enhanced FakeTensor's trace tooling serialization by storing symbolic storage size metadata alongside explicit optimization hints. Previously, when a storage size was a symbolic SymInt, trace tooling only stored the symbolic expression. This limited diagnostic and policy tools from accessing concrete expected extents provided by optimization hints. The new behavior separates concerns: the existing `size` field remains symbolic, preserving provenance for downstream trace consumers, while a new `size_hint` field is added only when every free symbol in the storage-size expression has an explicit hint override.

This design avoids specializing tensor shapes or changing runtime semantics, keeping the symbolic trace contract intact. The pull request (PR #183839, approved by ezyang and laithsakka) fixes issue #183835 and passes tests in `test_fake_tensor.py`. The change is critical for tools that rely on both the abstract symbolic representation and concrete guarantees for policy decisions, enabling more accurate resource planning and debugging in PyTorch's meta-tensor framework.