Feature: Time to First Result (TTFR) estimator integrated into agent pipeline

[rajnisk]

Relates to #26.

Summary

Integrates the TTFR (Time to First Result) estimator into the agent pipeline and exposes estimated retrieval time in chat response metadata. Tests are located under tests/ as requested.

Changes

• TTFR module (backend/ttfr_estimator.py): Core estimation logic with datasource defaults, modality inference, and three-phase breakdown (access, preprocessing, first output). Code-quality fixes applied (docstrings, rounding/singular forms, optional types).
• Tests (tests/test_ttfr_estimator.py): 39 unit tests; run with pytest tests/ -v from repo root. Pytest configured in pyproject.toml (pythonpath, testpaths); pytest remains a dev dependency only.
• Pipeline integration (backend/agents.py): TTFR computed per fused result in fuse_results; ttfr_estimates (id, title, ttfr_summary) stored per session and exposed via get_last_response_metadata(session_id).
• API (backend/main.py): Chat response metadata includes ttfr_estimates for the returned results.
• Retrieval (backend/retrieval.py): VertexRetriever.is_enabled implemented as a property for ABC compatibility (Python 3.14).
• Docker (backend/Dockerfile, backend/service-account.json.example): Build works without a real GCP key file by using an example placeholder.
• Script (scripts/verify_ttfr_m1_m2.sh): Local verification for M1 and M2 without GCP.

Verification

• cd backend && python3 -c "from ttfr_estimator import estimate_ttfr; print(estimate_ttfr(datasource_id='scr_005031_openneuro'))" and python3 demo_ttfr.py
• pytest tests/ -v (from repo root)
• POST /api/chat with a search query returns metadata.ttfr_estimates in the JSON response.

Checklist

• TTFR integrated into agent pipeline
• Estimated retrieval time included in chat response metadata
• Tests moved to tests/ directory
• No breaking changes to existing API
• Branch updated with upstream main

[rajnisk]

@QuantumByte-01 Would love your feedback when you have a moment.