Stream or batch SQLite exports to avoid full-result materialization

[tomanizer]

Problem

The SQLite export path materializes all rows in Python via fetchall() before writing them out. That is a memory-risky pattern and will scale poorly as export sizes increase.

Proposed change

• replace fetchall()-style SQLite export generation with batched fetch/write logic
• measure memory and latency impact using backend benchmarks or focused export tests

Likely files

• server/export_service.py
• export-related tests and benchmarks

Acceptance criteria

• SQLite export path does not require materializing the full result set in memory
• existing export API behavior remains correct
• tests cover large-ish result export behavior