Pipeline Resilience Phase 4: Add retry logic to periodic LLM scripts

[madjin]

Context

After Phases 1-3 harden the daily-critical scripts, 5 periodic/on-demand scripts still lack retry logic. They fail less often (monthly/quarterly runs) but the same failure modes apply.

Scope

Apply the same retry pattern (MAX_ATTEMPTS = 2, RETRY_DELAY_SECONDS = 5, completion token check, debug sidecar) to:

Script	API call line	Frequency
scripts/etl/helpers.py	~743	Monthly
scripts/etl/generate-monthly-retro.py	~303	Monthly
scripts/etl/generate-quarterly-summary.py	~200	Quarterly
scripts/etl/extract-entities.py	~262	Per-run
scripts/integrations/hackmd/update.py	~685	Daily

Same pattern each time: import time, retry constants, wrap API call in for attempt in range(MAX_ATTEMPTS) loop, token sanity check, debug sidecar on final failure.

Verification

Run a periodic script and verify [attempt 1/2] appears in logs and _metadata.attempts field in output.

Dependencies

• Phase 1 (issue Pipeline Resilience Phase 1: Add retry logic to daily critical LLM scripts #53) -- establishes the retry pattern to copy