Add agent flat-file context store and wire into onboarding + SIF agents

[AJaySi]

Motivation

• Provide a deterministic, ultra-fast read path for AI agents by persisting canonical onboarding snapshots to a per-user flat-file workspace to reduce latency for repetitive agent reads.
• Ensure agent tooling has a safe, isolated, and size-governed context layer with clear fallback to the database and SIF semantic index when flat files are missing or stale.

Description

• Add AgentFlatContextStore (services/intelligence/agent_flat_context.py) implementing atomic JSON writes, sensitive-key redaction, payload-size trimming, per-step save/load helpers, and a manifest index.
• Persist onboarding snapshots into the flat context from onboarding flows by updating StepManagementService to save Step 2 (website analysis), Step 3 (research preferences), Step 4 (persona), and Step 5 (integrations) to the flat store.
• Extend SIF agent surfaces by adding txtai agent tools and flat-file reader/writer hooks in services/intelligence/sif_agents.py, and add a fallback-aware retrieval API in services/intelligence/sif_integration.py to fetch step context via flat file -> database -> semantic index.
• Add comprehensive design and operational docs under docs/flat_file_context/ and a note in the SIF architecture docs describing the fast-context layer and retrieval order.

Testing

• Ran the backend unit test suite via pytest against modified modules and observed no regressions (all unit tests passed).
• Executed onboarding integration smoke tests exercising StepManagementService save flows for steps 2–5 and validated flat files and manifest were created and readable (smoke tests succeeded).
• Performed basic agent-tool smoke checks using SIFIntegrationService.get_step2_website_context and the txtai flat-context tools to confirm the flat-file -> DB -> semantic fallback behavior worked (smoke checks succeeded).

---

Codex Task