Batch the order & copy from prescribing_norm_tmp

[madwort]

• I tested this locally with a constrained environment:
  • 500MB RAM & 50GB tmp disk available to duckdb
  • just the 2026 data
• For the current main ingest:
  • tmp disk usage reached 2.5GB whilst creating the tmp table
  • tmp disk usage reached 6.5GB whilst inserting from the tmp table to the actual table
• For this branch ingest:
  • tmp disk usage reached 3.3GB whilst creating the tmp table
  • tmp disk upage remained at 3.3GB whilst inserting from the tmp table to the actual table
• It was a surprise that the tmp table appeared to require more tmp disk space to do the same thing, however the overall usage was significantly lower
• Wall clock time on my developer laptop was similar for this PR compared to main, however on the production server I would suspect that it would be quicker
  • this is because I think the tmp disk is on a slower hard-drive type device, and attached over a network, which I think it would make it relatively a lot more expensive than the storage on my developer laptop
  • the only real way to compare this is to experiment with it on production or a production-class machine