Mitigate SQLite write contention on import/correlation paths

[BadgerOps]

Summary

The backend currently performs write-heavy import/correlation work directly in request handlers against SQLite. This creates contention under concurrent usage and degrades API latency/reliability.

Why this matters

• SQLite has a single-writer bottleneck for this workload pattern.
• /api/imports/* and correlation routines can hold transactions long enough to impact other requests.
• As data volume/concurrency grow, this becomes an availability issue, not just a performance issue.

Repo evidence

• Import write path: backend/routers/imports.py
• Correlation merge path: backend/services/correlation.py
• DB setup/migrations: backend/database.py

Scope

• Add a background execution path (or equivalent chunked transactional strategy) for heavy ingest/correlation operations.
• Keep request handlers short-lived and return job status for long-running work.
• Add lock/latency instrumentation around write-heavy endpoints.

Acceptance criteria

• Import/correlation no longer run as long blocking DB writes on critical request paths.
• Concurrent import test demonstrates improved p95/p99 response times.
• Operational docs define expected limits and migration trigger points (SQLite -> Postgres).

Out of scope

• Full DB engine migration in this issue.