Improve the `pip.parse` API to allow for incremental building of the configuration

[aignas]

Currently the API for defining different parameters by target platform is not ideal and we need to jump over hoops whilst trying to maintain it. The main difficulties are:

• We would like to specify different pip.parse attributes per target platform, but we cannot, because some of the parameters need to be labels, label lists or other.
• We cannot have overrides for these things easily and the definitions of what those target platforms are are hard-coded in the code.

The idea that I have is to reuse the same recipe from #2578, where we create a builder for the configuration and it allows us easily define:

• default constraint and flag values for the config settings that we create for different platform variants.
• default values for the pip index url, etc
• default values for which platforms we should support.
• default auth configuration for the bazel_downloader.
• The list of python_versions that are supported pip.parse(python_version=) should not be mandatory #1708.
• build the configuration incrementally for the pip.parse invocation allowing per-target-platform configuration of:
  • extra_pip_args. feat: add support for passing extra pip args by platform #2745
  • requirements_lock.
  • constraint_values and flag_values. pip_parse: requirements_by_platform: allow custom defined platforms #2548

What is more the override API could be blended in more easily to provide better support for specifying different, patches, etc.

[aignas]

Taking the example from #2578 (comment)

We should have something like:

pip.defaults(
    hub_name = `pypi`,
    index_url = "https://pypi.org/simple/", # once we want to enable the `experimental_index_url` for everyone or all pip.parse repos regardless where they come from.
    index_url_overrides = {},
)
pip.defaults(
    platform = "cp39_linux_x86_64",
    constraint_values = ["@platforms//os:linux", "@platforms//cpu:x86_64"],
    flag_values = {
        "//python/config_settings:python_version_major_minor": "3.9",
    }
)
pip.defaults(
    platform = "cp313t_linux_x86_64",
    constraint_values = ["@platforms//os:linux", "@platforms//cpu:x86_64"],
    flag_values = {
        "//python/config_settings:python_version_major_minor": "3.13",
        "//python/config_settings:py_freethreaded": "yes",
    }
)
pip.defaults(
    platform = "cp313_linux_x86_64",
    constraint_values = ["@platforms//os:linux", "@platforms//cpu:x86_64"],
    flag_values = {
        "//python/config_settings:python_version_major_minor": "3.13",
        "//python/config_settings:py_freethreaded": "no",
    }
)
# At user site
pip.configure(
    name = "pypi",
    pip_args = ["default"],
)
pip.configure(
    name = "pypi",
    requirements = "//:requirements_darwin.txt",
    platform = "cp313t_osx_x86_64",
    pip_args = ["only for this platform"],
)
pip.configure(
    name = "pypi",
    requirements = "//:requirements_linux_gpu.txt",
    platform = "cp313gpu_linux_x86_64",
    pip_args = ["only for this platform"],
    constraint_values = ["@platforms//os:linux", "@platforms//cpu:x86_64"],
    flag_values = {
        "//my_custom_flag": "cuda",
        "@rules_python//python/config_settings:python_version_major_minor": "3.13",
        "@rules_python//python/config_settings:py_freethreaded": "no",
    },
)
pip.configure(
    name = "pypi",
    requirements = "//:requirements_linux_3_10_11_workaround.txt",
    platform = "cp310.11_linux_x86_64",
    pip_args = ["only for this platform"],
    constraint_values = ["@platforms//os:linux", "@platforms//cpu:x86_64"],
    flag_values = {
        "@rules_python//python/config_settings:python_version_major_minor": "3.10.11",
    },
)

I am not sure if all of these permutations are really necessary, but the whole override idea is really nice.

Whilst the uv the uv_version is the main key where the configuration is built/modified. Here we could have the hub_name.

OK, that is it for now.