Preserve historical leaks data across post-processor runs

[dmarulli]

Summary

• The leaks post-processor currently uses CREATE OR REPLACE TABLE to rebuild the leaks detail table on every run, wiping all data outside the 30-day computation window
• This replaces that with a DELETE + INSERT scoped to the window, so historical leak detection results are preserved
• The agg table still rebuilds from the full detail table each run (now includes full history)
• Adds CREATE TABLE IF NOT EXISTS as an initialization step for new orgs that won't have a leaks table yet to delete from or insert into

Known issue: leaks_cadc_south_tahoe

This org's leaks table has an older schema (e.g. FLOWTIME instead of FLOWTIME_TS) from before column aliases were added to the leaks detection query in snowflake.py. The South Tahoe DAG is not currently running, so this is not a problem. If it's ever reactivated, the old table should be dropped first so CREATE TABLE IF NOT EXISTS can recreate it with the current schema.