Documentation: Update assessment contract

Problem

The assessment architecture docs under docs/architecture/assessment/ still describe the OLD BATCH contract (request query template, input_schema nested under assessment.params, a duplicate_detection pre-filter). After the config/input reshape (#1154, #1153) these are stale and mislead readers.

Scope

Update the following to the new contract — prose, JSON examples, and tables. Do not touch the .png assets.

• docs/architecture/assessment/README.md
• docs/architecture/assessment/api-contract.md
• docs/architecture/assessment/configuration-and-versioning.md
• docs/architecture/kaapi-ai-assessment-ARCHITECTURE.md

Contract changes to reflect

1. query removed from the request input. BATCH input = { "data": [...] } (detected by the data key). RESPONSE input carries only attachments and still returns 501.
2. Prompt template moves into the config as submission — mandatory on assessment.params.submission, optional per pre-filter's params.submission. Interpolated server-side per row.
3. input_schema hoisted to the config_blob top level (sibling of pre_filters/assessment), mandatory and non-empty — no longer under assessment.params.
4. duplicate_detection pre-filter removed entirely (incl. knowledge_base_id and the output.pre_filter.duplicate_detection result field). Only topic_relevance remains.
5. Placeholder validation now at config-save (every {column} in a submission must resolve against the top-level input_schema); runtime validates only the structured data rows.

Acceptance criteria

• No doc describes a request query field or a duplicate_detection pre-filter.
• All JSON examples show top-level input_schema, submission in the config, and data-only request input.
• Config-save vs runtime validation split documented.
• .png assets untouched.

Related: #1153, #1154.