chore: start explainable fraud holds

[jenetbamidele2-ai]

Summary:

Imlement explainable fraud holds without exposing sensitive fraud-detection signals.

This change will:

• Add a safe, deterministic sxplanation model for fraud holds.
• Keep raw signals such as addresses, funders, transaction hashes, graph edges and pattern keys out of user-facing responses.
• Preserve admin authorization, scope checks, validation, rate limiting and correlation IDs.
• Maintain idempotent and concurrency-safe fraud-flag updates without resetting reviewer state.
• Add focused coverage for successful scans, invalid input, duplicate/retry behaviour, boundary cases, authorization, redaction, and failure recovery.
• Update the API contract and operational documentation.

Security and failure mode handling:

• Explanations expose only approves reason codes and user-safe descriptions.
• Raw detection evidence remains restricted to the internal review boundary.
• Invalid requests are rejected deterministically.
• Duplicate and concurrent scans converge through atomic upserts.
• Detector failure remain diagnosable through correlation IDs and structured legs without logginf sensitive signals.
• Existing callers remain compatible through additive response changes and preserved repository interfaces.

Validation

• Focused fraud service, worker, and admin route tests.
• Authorization and scope regression tests.
• Redaction and response-contract tests.
• Retry, duplicate, concurrency and partial-failure scenerios.
• Repository CI checks.

Closes #933

[greatest0fallt1me]

This PR is linked to issue #933, but it currently contains no changed files, so there is no implementation to review. Please add the substantive code and focused tests required by the issue; I’m leaving the PR open for that update.