fix(codes): reconcile time-zone table with USGS and WQX code lists

[thodson-usgs]

Follows up #455 (merged); rebased onto main as a single commit.

Summary

codes.timezones turns the abbreviation beside a local time into the UTC offset used for the *DateTime columns that waterdata.get_samples() and wqp.get_results() append. The table now follows the code lists those services use: the USGS NWIS time-zone list (Samples) and the WQX TimeZone domain (WQP).

• American Samoa in Samples: Samples writes this zone as ZP-11, which the table lacked, so every timed American Samoa row came back with a NaT datetime. It is now UTC-11.
• Codes that used to become NaT: 39 more codes from the two lists now resolve: the other ZP* offsets, IDLW/IDLE, NZT, SWT, the retired WQX codes AHST and YST, and others.
• Behavior change: codes the table read differently from the USGS list now use the USGS meaning: CAST +9:30, CCT +8, EASST +11, EAST +10, NFT -3:30, NZDT +13, WAST +7, WAT -1. FJST moves to +13 too.
• Where the two lists conflict, WQX wins: SST (Samoa in WQX, Swedish Summer Time in USGS) stays UTC-11, the meaning WQP returns. BST becomes UTC-11. Samples doesn't use either code for Samoa; it writes ZP-11.
• Snapshots: copies of both lists are in tests/data/, and the module docstring points to them.

Tests

• tests/timezones_test.py, offline: checks that every code in each list maps to that list's offset, and that WQX_WINS == {"SST", "BST"} covers every conflict between the lists, so any new conflict has to be decided explicitly.
• tests/timezones_test.py, live: checks that the published lists still match the snapshots, and that Samples/WQP responses for Pacific monitoring locations have no NaT among timed rows.
• utils_test.py: an end-to-end ZP-11 case through _attach_datetime_columns.
• With the old table, 50 of the new tests fail; with the new one, all pass.
• Gates: ruff, mypy, lint-imports, xenon, complexipy; 1757 tests pass; coverage 99.08%.

Live verification (2026-10-16)

Service	Monitoring location / query	Code	Timed rows with NaT
Samples	USGS-141037169391201 (Am. Samoa), 566 rows	ZP-11	0 (all were NaT before)
Samples	USGS-16854500 (Guam), 969 timed	GST	0
WQP legacy	statecode=US:60, since 2023	SST	0 of 591
WQP legacy	statecode=US:66, since 2023	GST	0 of 1099
WQP WQX3	11NPSWRD_WQX-NPSA_F_07TUT_FW	SST	0 of 846
WQP WQX3	USGS-16854500	GST	0 of 969

Example: Samples 2011-09-09 09:40 ZP-11 → 2011-09-09 20:40Z.

Not done

• Silent NaT: an unrecognized code still becomes NaT without any warning.
• One table for both services: splitting into a separate table per service is deferred until a USGS response is seen using SST or BST with its USGS meaning.

Stress test (2026-10-16)

wqp.what_activities plus _attach_datetime_columns and waterdata.get_samples(service="activities") for all 50 states, DC, and five territories, over four windows (Jan–Mar 1995, Jul 2010, Jun 2024, Jan 2025): about 520,000 timed rows and 25 distinct codes.

• No timed row became NaT.
• Every resolved offset falls within ±1 h of its state's zones, apart from UTC/GMT stamps, two Alaska rows stamped AST/ADT, and one Idaho well (USGS-433812111574701) whose ten activities are all stamped WAST. That code is wrong in the source record whatever the table does; the old table put it at +2, the USGS list at +7.
• WQP passes USGS codes through: ZP-11 appears in WQP results for USGS American Samoa records, so this fix reaches wqp.get_results() too. A live case now covers it, and the docstring and NEWS say so.