[Bazel] Add Bazel rules for Python bindings

[kalvdans]

Summary

Adds Bazel build support for the torch-mlir Python bindings. Until now the
Bazel build (utils/bazel/) only covered torch-mlir-opt and the C++
libraries; the Python package could only be built via CMake. This brings the
Bazel build closer to parity with CMake for projects that consume torch-mlir
through Bazel.

What's included

• C API libraries in torch-mlir-overlay/BUILD.bazel: CAPITorch and
  CAPITorchRegisterEverything (via mlir_c_api_cc_library), plus an
  exports_files for include/torch-mlir-c/Registration.h.
• python/BUILD.bazel (new):
  • generates the Torch op bindings (_torch_ops_gen.py) with
    gentbl_filegroup;
  • builds the _torchMlir.so (pybind11) and _mlirRegisterEverything.so
    (nanobind) extension modules;
  • assembles everything into a torch-mlir py_library, mirroring the
    package tree CMake produces (torch_mlir/, torch_mlir/dialects,
    torch_mlir/_mlir_libs).
• python/fileutils.bzl (new): a small copy_into_directory rule that
  symlinks generated files into that package layout.

Notes

• Per the Bazel support policy,
  this is peripheral-tier and opt-in. The default Bazel CI target
  (@torch-mlir//:torch-mlir-opt) is unaffected.
• Like MLIR's own Bazel Python bindings, building the Python targets requires
  the consuming project to provide nanobind and a Python toolchain
  (@rules_python//python/cc libs) through its module graph; torch-mlir's
  standalone Bazel build does not register these.
• BUILD/.bzl files were formatted with buildifier as requested by the
  contributing guidelines.

Testing

• bazel query @torch-mlir//python:torch-mlir + @torch-mlir//:CAPITorch + @torch-mlir//:CAPITorchRegisterEverything resolves.
• Analysis (bazel build --nobuild) of @torch-mlir//python:torch-mlir
  configures the full dependency graph against the pinned LLVM revision.
• The targets build and import as a py_library dependency in a downstream
  Bazel (bzlmod) project that provides the nanobind/Python-toolchain deps.