feat: support custom CLI args for Python config modules

[eric-tramel]

Summary

When using a Python config module (.py file) with data-designer create or data-designer preview, there is currently no way to pass custom arguments to the user's load_config_builder() function. This limits the flexibility of custom config scripts — users must hardcode values or manage their own env-var-based workarounds.

Proposed behavior

Allow users to pass extra arguments after a -- separator, which get forwarded to the Python module's load_config_builder() function:

data-designer create ./my_custom.py --num-records 32 -- --key value --key2 value2
data-designer preview ./my_custom.py -n 10 -- --model large --temperature 0.7

Everything after -- would be captured and passed to load_config_builder as a context object (e.g. a list[str] or a lightweight wrapper). The user's function signature would then optionally accept this:

# Before (still works, no args)
def load_config_builder() -> DataDesignerConfigBuilder:
    ...

# After (opt-in to receive CLI args)
def load_config_builder(cli_args: list[str] | None = None) -> DataDesignerConfigBuilder:
    builder = DataDesignerConfigBuilder()
    if cli_args:
        # User parses as they see fit (argparse, simple iteration, etc.)
        ...
    return builder

Changes required

1. CLI commands (create, preview, validate): Accept and capture arguments after -- via Typer's context.args or Click's allow_extra_args mechanism.
2. load_config_builder() in config_loader.py: Accept an optional cli_args: list[str] | None = None parameter and pass it through to _load_from_python_module.
3. _load_from_python_module(): Inspect the user's load_config_builder signature. If it accepts a parameter (via inspect.signature), pass cli_args; otherwise call with no arguments for backwards compatibility.
4. GenerationController._load_config(): Thread the extra args from the command through to load_config_builder().

Why

Custom Python config modules are a key extensibility surface. Users frequently want to parameterize their configs — e.g. switching model endpoints, adjusting generation parameters, or toggling column sets — without maintaining separate config files for each variant. Passthrough CLI args are the standard Unix pattern for this and would make the Python module path significantly more useful.

Backwards compatibility

• YAML/JSON configs: unaffected (extra args after -- are simply ignored or produce a clear error).
• Existing Python modules with def load_config_builder(): (no params): continue to work as-is via signature inspection.
• Only Python modules that opt in by adding a cli_args parameter will receive the extra arguments.