refactor(cli): unify -m/--model option across all commands

[xieofxie]

@

Summary

Unifies the -m/--model option across every command in src/winml/modelkit/commands/ by routing them all through the shared model_option / model_path_option helpers in utils/cli.py, instead of ad-hoc inline click.option definitions. The flag spelling, ordering, type, and validation now have a single source of truth.

Closes #563.

Changes

Helpers (utils/cli.py)

• model_path_option (validated ONNX Path) — added multiple and help_text parameters.
• model_option (any model reference, no validation) — added multiple and help_text parameters.

Commands converted from inline click.option to the shared helpers

Command	Now uses
compile	model_path_option(required=False, multiple=True, help_text=…)
optimize	model_path_option(required=False, help_text=…)
quantize	model_path_option(required=True, help_text=…)
export	model_option(required=True, help_text=…)
eval	model_option(required=False, multiple=True, help_text=…)
build	model_option(required=False, help_text=…)
inspect	model_option(required=False, help_text=…)

(analyze, run, serve, perf, config already used the helpers.)

Destination standardized to model — build and inspect previously bound --model to model_id; the CLI binding is now model. Downstream keyword args that are genuinely the domain concept (e.g. WinMLQuantizationConfig.model_id) are unchanged.

This is purely structural — each command keeps its existing accept-types (PATH-validated vs flexible TEXT), required semantics, multiple, and help wording.
@