deps: raise torch upper bound to <2.13

`pyproject.toml`'s `torch>=2.8,<2.12` cap was older than what the CI
build pipeline now picks up. `env/build_requirements.txt` and
`env/test_requirements.txt` leave `torch` unpinned, so `uv pip install
--extra-index-url https://download.pytorch.org/whl/cu130 ...` resolved
it to whatever was newest on the CUDA index.

That worked while the newest cu130 torch matched the wheel's range.
Once torch `2.12.0+cu130` was published, the build env installed
2.12.0+cu130 (so fvdb's `.so` ended up compiled against the 2.12 ABI,
which includes `at::TensorBase::const_data_ptr<int8_t>`), but the
*test* env's install of the resulting wheel hit pyproject's `<2.12`
ceiling and pip downgraded torch to the CPU-only `2.11.0` from PyPI
(which is missing those CUDA-side template symbols). The CUDA 13.0
unit-test pipeline then surfaces:

    ImportError: .../libfvdb.so: undefined symbol:
      _ZNK2at10TensorBase14const_data_ptrIaEEPKT_v

at `import fvdb` time. Every CUDA 13.0 PR since 2026-05-16 has been
red on this; CUDA 12.8 isn't affected (its index still serves a
wheel in-range).

Bumping the upper bound to `<2.13` lets the wheel and both the build
and test environments converge on the same torch 2.12.x build, with
no extra pin needed in the requirements files. A short comment in
pyproject.toml documents the convention so the next torch minor
release bump is mechanical.

Signed-off-by: Francis Williams <francis@fwilliams.info>

Author: fwilliams

pyproject.toml

@@ -11,8 +11,13 @@ license = { text = "Apache-2.0" }
 requires-python = ">=3.10"
 readme = "README.md"
 dependencies = [
-    # Require torch; users will choose the correct CUDA build from PyTorch's index
-    "torch>=2.8,<2.12",
+    # Require torch; users will choose the correct CUDA build from PyTorch's index.
+    # Upper bound is one minor release ahead of the highest torch we currently
+    # build against in CI; bumping the upper bound here is what `env/{build,test}_
+    # requirements.txt` rely on (both are unpinned, so `uv pip install` follows
+    # the wheel's range). When the build / test pipelines pick up a newer torch
+    # than this range, bump both.
+    "torch>=2.8,<2.13",
     "numpy",
 ]
 optional-dependencies = {viewer = ["nanovdb-editor>=0.0.23,<0.2.0"]}