[Quanwritescodes]

Summary

This PR adds the historical replay and multi-network isolation foundations required for deterministic audits and conflict resolution in the backend.

What changed

• Added the snapshot pipeline under backend/src/snapshot/ with a canonical, sorted aggregation path to make the generated output deterministic.
• Added a pinned historical model-version registry so the replay logic uses a fixed model version for a given historical range instead of drifting with the current production model.
• Added the replay entry point under backend/src/replay/ to regenerate snapshots from raw ingested rows for historical ranges.
• Added explicit network identity handling under backend/src/network/ with Mainnet | Testnet | Futurenet and schema validation that fails fast if a table is missing a network discriminator.
• Added regression tests for repeated replay stability and network isolation to guard against cross-contamination and ordering-dependent output.

Why this matters

• Audit and dispute scenarios depend on being able to regenerate the same historical snapshot from the same raw data with byte-identical output.
• Multi-network ingestion must never mix mainnet, testnet, or futurenet data in shared caches, topics, or persisted analytics.
• This reduces the risk of silent data contamination that is difficult to detect after the fact.

Validation

• Added deterministic replay tests that exercise repeated historical range replays.
• Added dual-network ingestion tests to confirm zero cross-contamination between networks.
• Added startup schema validation to reject tables lacking the required network field.

Notes

• The work is intentionally designed to be deterministic from the historical replay boundary onward and to keep network identity explicit at every persistence and cache layer.
• This change establishes the deterministic replay and isolation foundation for subsequent production hardening and telemetry work.

closes #325
closes #326