Upgrade Bzlmod dependencies and improve NumPy 2.0 casting compatibility

[copybara-service]

Upgrade Bzlmod dependencies and improve NumPy 2.0 casting compatibility

This CL introduces several Bzlmod dependency upgrades and core runtime/toolchain improvements:

1. Bzlmod Dependency Alignments:

   • Upgrade @rules_python from 1.0.0 to 2.0.1 and @rules_cc from 0.0.9 to 0.2.17 in both MODULE.bazel and examples/MODULE.bazel to align toolchains across common packages.
   • Declare aspect-based dependencies (rules_cc and grpc) in MODULE.bazel to support standard Bzlmod proto loaders.

2. Robust, re-free NumPy 2.0 Casting Checks:

   • Upstream standard warnings context manager and ValueError/TypeError exceptions catching inside dtype_utils.py to handle NumPy 2.0 casting overflow check differences (adopting NEP 50).
   • Integrate a robust, re-free string version parser (_parse_version_string) using basic string splitting (split('.')) and digit extraction. This completely removes the regex import re dependency and safely handles pre-releases (e.g. '2.0rc1'), dev builds (e.g. '2.0.0.dev0'), and incomplete version strings without runtime crashes.
   • Expose structure target visibility directly as public inside common_libs/BUILD to satisfy Bazel's strict dependencies checks for downstream consumers.

3. Encapsulate Dev Requirements Tooling:

   • Move toolchain-dependent compile_pip_requirements targets and @python loads out of the root workspace.BUILD.bazel to a dedicated subpackage tools/requirements/BUILD.
   • This completely prevents Bzlmod transitive loading and analysis failures for external downstream consumers, while fully preserving lock-file regeneration capabilities for developers.
   • Export requirements lock files in workspace.BUILD.bazel via exports_files to resolve subpackage visibility rules.