Apply PUE and WUE for the year the usage was incurred

[jnioche]

Fixes #249

AWS publishes PUE and WUE per year, and the figures move noticeably from one year to the next — eu-central-1 went 1.32 → 1.33 → 1.35 → 1.24 across 2022-2025, and its WUE went from 0.01 in 2024 to 0.17 in 2025. Applying a single blanket value skews any report spanning more than the latest year.

What changed

aws-pue-wue.csv is now Geography,RegionID,Year,PUE,WUE with one row per region per year, 2022-2025.

The figures come from the tables on the AWS sustainability page rather than the downloadable aws-wue-pue.csv, which was last updated in December 2025 and stops at 2024. Using the page picks up:

• the 2025 figures;
• AWS's restatements of earlier years (e.g. eu-south-2 2024 is 1.07 on the page, 1.09 in the stale CSV);
• the regions AWS has started reporting on since — London, Paris, Tel Aviv, Bangkok, Osaka — plus PUE for Sydney and Melbourne, which previously had WUE only.

The one thing the page does not carry is the global WUE for 2022 and 2023; those two values are taken from the downloadable CSV, which is noted in the file header.

PWUE keys its lookups by region and year:

1. exact region id
2. geography regex (eu-.+, (me|il)-.+, (us|ca|mx)-.+, …)
3. the new GLOBAL row
4. the configured default

Within a tier it takes the row for the usage year, falling back to the closest year published for that region. That fallback is what covers recently-reported regions, WUE that only exists from 2024 onwards, and rows with no usable date.

Reading the year: new RowColumn.getYear(), probing ChargePeriodStart → line_item_usage_start_date → Date → BILLING_PERIOD. All four are read optionally, so the module keeps working across AWS/Azure and NATIVE/FOCUS without adding a required column to columnsNeeded(). It handles Spark timestamps and dates as well as the string forms CSV exports use. LINE_ITEM_USAGE_START_DATE/_END_DATE moved from the AWS FOCUS bridge into CURColumn so the shared module can reach them.

azure-pue-wue.csv takes the same schema with the year left empty — meaning the row applies to every year — since Microsoft does not break its figures down by year.

Notes for reviewers

• Two behaviour changes. An unknown region now gets the global average for its year (1.14 PUE / 0.12 WUE in 2025) instead of the hardcoded 1.15 with no WUE. As a result the default config key is now a genuine last resort, reachable only for Azure, whose CSV has no GLOBAL row.
• Regex entries moved from a HashMap<Pattern, Double> to an ordered list. Pattern has identity equality, so once a key carries several years the map would have split one regex across several entries; the list also makes the match order deterministic.
• il-central-1 used to have a standalone row carrying the Middle East average. It now has its own 2025 figure and falls back to the (me|il)-.+ geography average for earlier years.
• RowColumnTest is new rather than an addition to ColumnTest, because ColumnTest, CURColumnTest and AzureColumnTest are written against JUnit 4 and the project only ships the Jupiter engine — Surefire skips all three silently. Worth fixing, but separately from this change.

mvn test: 362 tests, all passing.