FIX: Apply the request limits to the remaining request fields

[varunj-msft]

Description

#2960 added request limits, but some fields didn't get them: fields added since (conversion markers in #3002, technique creation in #3047, seed browsing in #2989) and fields from PRs that landed while #2960 was open (the scorer API in #2990, async message sends in #2861, conversation save in #2853). Separately, #2960 declared conversation_id on GET /api/attacks/{attack_result_id}/messages as IdentifierStr = Query(...), and FastAPI drops the length limit of an annotated alias on a required query parameter, so any length reached the service.

These now return 422 above the existing limits:

• start_token and end_token on add-message and converter preview: identifier length.
• POST /api/techniques: params keys, converter names, and adversarial_chat (identifier length), and description (text length). The adversarial prompts stay limited only by the body size, like other prompt content.
• GET /api/datasets/seeds and /seeds/{example_id}: selection_key (cursor length) and harm_category values (label value length).
• Message-send path ids.
• Scorers: type, params keys and count, cursor, and scorer_registry_name, matching targets and converters.
• target_registry_name on POST /api/attacks/save-conversation.
• conversation_id: the limit is set on the Query itself.

Not included: GET /api/media?path (bounded by the URL length limit) and the label filters (checked by their validator).

Tests and Documentation

• test_common_models.py, test_api_routes.py, test_message_send_routes.py, test_scorer_routes.py, test_seed_example_routes.py: each added limit rejects one character or item over, and the routes return 422 before the service is called, including an over-long conversation_id.
• test_common_models.py also checks that technique fields accept values at the limit while the adversarial prompts stay unbounded.
  No JupyText changes.