Feat/humanitarian verification v2 parsing

[Opulencechuks]

Closes #269

Summary

This PR unifies the parsing logic for humanitarian verification responses across OpenAI, Groq, and the Test provider. It introduces a version 2 schema (HumanitarianVerificationResponseV2) to ensure that callers receive stable, identical-shape responses across all provider pathways, resolving the issue where responses built out-of-band or truncated mid-token could break verification endpoint expectations.

• Unified Parsing: Added parse_verification_response and _normalize_verification_dict in humanitarian_verification.py to handle markdown block stripping, missing keys padding, and regex-based recovery parsing for mid-token truncated JSON responses.
• Metadata Stamp: Every response now includes a deterministic stamp dictionary referencing the provider, model, and prompt_variant.
• V2 Schemas: Created app/ai-service/schemas/humanitarian_verification_v2.py implementing HumanitarianVerificationResponseV2 and HumanitarianVerificationDetailsV2.

Testing

Tests were added and verified to ensure compatibility and schema correctness across all three providers.

# Run schema and verification tests
pytest app/ai-service/tests/test_schemas.py -v
pytest app/ai-service/tests/test_humanitarian_verification.py -v

Screenshots

(No UI changes; backend service APIs and schemas update only.)

Checklist

• Tests added or updated for new/changed behavior
• Existing tests pass
• No secrets, keys, or seed phrases committed
• Follows the coding conventions in the relevant service README
• PR is focused on a single concern

[kilodesodiq-arch]

Awesome work on this! Thanks for the contribution 🚀 Merging now.