fix(cli): remove --evaluators, which never affected a run

[kmadan]

🚀 Overview

Removes --evaluators from the evaluate subcommand. The flag was accepted and
had no effect on any run. Follow-up to #90, which removed it from the documented
options.

🔍 Why it never worked

aicertify_app_for_policy derives the evaluator set from the RequiredMetrics
the selected policies declare:

# aicertify/api/policy.py:103
evaluators = evaluator_registry.discover_evaluators(metrics)

_run_evaluate accepted an evaluators argument and did not forward it:

return await aicertify_app_for_policy(
    contract=contract, policy_folder=policy_folder,
    output_dir=output_dir, report_format=report_format,
    custom_params=custom_params,      # evaluators not passed
)

The parser accepted the flag, argparse bound it, and the run proceeded
identically.

🎯 Why removed rather than wired through

Restricting the evaluator set below what a policy requires leaves that policy's
metric absent from the OPA input. An absent metric is not a failed one: a
rule that reads it goes undefined and its default stands.

A flag whose plain meaning is "run fewer evaluators" would therefore convert a
policy that should deny into one reporting no obstacle. That is a worse outcome
than the flag doing nothing, and it is the same fail-open class GOPAL's
empty-input test gate exists to catch.

Explicit evaluator selection remains available on evaluate_by_policy
(aicertify/api/policy.py:297) for callers using the API directly.

🎯 Why removed rather than warned

Removing the registration makes the change visible at the point of use:

$ aicertify evaluate --contract c.json --policy eu_ai_act --evaluators Fairness
aicertify: error: unrecognized arguments: --evaluators Fairness
$ echo $?
2

Nothing depended on the flag's effect, because it had none. Only its acceptance
changes.

📜 Changes Made

• ✅ Dropped the evaluators parameter from _run_evaluate (dead)
• ✅ Dropped evaluators=args.evaluators at the call site (dead)
• ✅ Dropped the ev.add_argument("--evaluators", ...) registration

Three deletions, 7 lines. No other behaviour changes.

🎯 How to Test

aicertify evaluate --help | grep -c -- --evaluators     # -> 0
aicertify evaluate --contract c.json --policy eu_ai_act --evaluators X; echo $?   # -> 2

📄 Notes for reviewers

Every flag the parser registers was audited for the same defect. All sixteen
others have a destination that is read somewhere in cli.py; --evaluators was
the only one that did not. --with-llm-metrics on the demo subcommand, its
nearest neighbour, is genuinely wired (cli.py:112).