ExportWorker holds Postgres connection across chunked export

Context

A single Lightning.WorkOrders.ExportWorker job failed in the last 90 days
with a DBConnection.ConnectionError ("tcp recv: closed"):

(lightning) lib/lightning/workorders/export_worker.ex:175:
    Lightning.WorkOrders.ExportWorker.process_logs_and_dataclips_chunk/2

The export args show a ~3-month window with no extra filters — likely a large
result set. The pool closed the connection while the worker was mid-stream.

Root smell

process_logs_and_dataclips_chunk/2 (line 175) processes the export via a
single Stream.chunk_while that holds a Postgres connection across each
chunk's disk-write work. Long workorder lists + slow IO = connection held past
the pool timeout.

Options to evaluate

• Smaller chunks
• Dedicated checkout (Repo.checkout/2)
• Paginated export that releases the connection between pages
• Hard cap on export size with a user-facing message

Priority

Low — 1 incident in 90 days. Filing for tracking; revisit on recurrence or as
part of broader history-export improvements.

Note

The original issue body also called out two Sentry events for this single
incident. That turned out to be a generic pattern affecting every non-AI-queue
Oban error (Logger.error + explicit capture_exception), not an
ExportWorker-specific bug — it has been moved to #4747.

Start by reading lib/lightning/workorders/export_worker.ex at process_logs_and_dataclips_chunk/2, then trace how Stream.chunk_while handles database access and disk writes. Evaluate the listed options and define a bounded export approach that does not hold a Postgres connection across slow chunk work, with a user-facing size limit if that option is chosen.