feat(toolchains): add toolchain_target_settings to python.override

[novas0x2a]

Users who register custom toolchain families (e.g. gated behind a
string_flag like "custom" vs "prebuilt") have no way to prevent the
default-registered toolchains from acting as a silent fallback. When a
user requests a custom toolchain at a version they didn't register,
Bazel's toolchain resolution quietly falls back to the default prebuilt
toolchain instead of producing an error.

This adds a toolchain_target_settings attribute to python.override
that appends the given config_setting labels to the target_settings
of every toolchain registered by the module extension:

python.override(
    toolchain_target_settings = ["@@//:python_toolchain_family_prebuilt"],
)

These settings are appended to the target_settings of all toolchains
registered by the extension, including any that already have settings
from python.single_version_platform_override. Toolchains registered
outside the extension (e.g. via local_runtime_toolchains_repo) are not
affected.

Note: toolchain_target_settings is popped from the config default
dict before it reaches python_register_toolchains(), so it doesn't
leak into the python_repository kwargs.

Integration test passes across the full Bazel version matrix (7.7.0,
8.5.1, 9.0.0rc1). Verified the test fails without the fix (default
toolchain silently resolves as fallback).

Fixes #3673

[gemini-code-assist]

Code Review

This pull request introduces a toolchain_target_settings attribute to the python.override module extension tag, enabling users to globally append config_setting labels to the target_settings of all toolchains registered by the extension. This feature is particularly useful for gating default toolchains behind custom build settings to prevent unintended fallbacks. The PR includes the necessary logic in the extension implementation, updated documentation, and new integration and unit tests. Feedback was provided to clarify the documentation regarding how these global settings interact with toolchains that already have specific target_settings defined, ensuring users understand that the settings are appended rather than ignored.