Add Xylem Data Lake adapter

[dmarulli]

Summary

• New adapter (xylem_datalake) for fetching AMI data from Xylem's Data Lake platform via the Superset SQL Lab API
• Extracts from three tables: account (metadata), water_intervals (interval reads), water_registers (register reads)
• Auth via Keycloak PKCE OAuth with automatic session management, CSRF refresh, and re-auth with exponential backoff on expiry
• Parameterized by agency_code — works for any Xylem Data Lake agency (tested with Hillsborough)
• Auth module is designed to be swappable when Xylem rolls out MFA requirements

Files changed

• amiadapters/adapters/xylem_datalake.py — adapter implementation (extract, transform, raw loaders)
• amiadapters/configuration/models.py — config and secrets models, enum registration
• amiadapters/config.py — factory wiring
• docs/adapters/xylem_datalake.md — adapter documentation
• test/fixtures/xylem-datalake-*.yaml — local test config fixtures
• test/manual/test_xylem_datalake.py — manual end-to-end test script

Test plan

• Extract tested against Hillsborough Data Lake: 5,147 accounts, 87K interval reads, 87K register reads
• Transform verified: 5,145 meters, 87K reads with correct field mapping
• Session expiry mid-extract handled gracefully (504 → re-auth → resumed)
• Snowflake load (uses shared load infrastructure, not separately tested)

[mdowell12]

Wow! Looks great @dmarulli . I hope it wasn't too painful to add the new adapter and I'm encouraged that you made one that looks this good on your first try :)

I saw your manual test. Would you consider adding unit tests for the adapter, too? Even some basic tests, especially of the transform, can be useful down the road. There are lots of examples in the codebase and I've found that AI tools are pretty good at this. Up to you.

To fix the linter, you can run black . in the root directory.

[dmarulli]

Thank you @mdowell12 ! It wasn't too painful. We did it in a few stages, with a bunch of AI support along the way. @christophertull had come up with the basic concept in a notebook and done some prototyping, I worked with that to polish some auth issues and wire up to snowflake, and then once we saw the prototype was generating clean data, I ported it into the adapter framework. I'm sure all the existing docs and transcript from our zoom walkthrough provided valuable context!

Merging.