CI Tests for Node 18 and 19

[justingrant]

Makes CI tests pass in Node 18 and Node 19. It was a little more complicated than https://github.com/tc39/proposal-temporal/pull/2448 because TS code and demitasse tests required some changes.

• Update CI scripts to test Node 19, 18, 16, and 14 using GH's "matrix" strategy.
• Add non-ISO-calendar snapshot data for Node 18/19
• Make demitasse tests resilient to format changes in CLDR 42 (which is included in ICU 72 which is used in GH Actions' deployments of Node 19)
• Switch to use the optimized Test262 runner from @js-temporal/temporal-test262-runner
• Snap to the latest Test262, which contains PRs that make those tests work with CLDR 42:
  • https://github.com/tc39/test262/pull/3750
  • https://github.com/tc39/test262/pull/3751
  • https://github.com/tc39/test262/pull/3762
• Adds a new test/expected-failures-todo-migrated-code.txt file that tracks Test262 failures due to this repo being behind proposal-temporal in merging the latest normative changes to the Temporal spec. As upstream changes are migrated here, tests should be removed from that file.

[justingrant]

All Demitasse tests now pass. Node 19 made many changes in the output of Intl.DateTimeFormat. It took a bit of time to write node-version-agnostic code for affected tests.

But I'm stumped with Test262, which is failing in Node 19 but when I look at the test output I don't see any failures, only successes and expected failures. @ptomato , you're a Test262 expert. Any ideas about what may be going on?

[justingrant]

But I'm stumped with Test262, which is failing in Node 19 but when I look at the test output I don't see any failures, only successes and expected failures. @ptomato , you're a Test262 expert. Any ideas about what may be going on?

Mystery solved by copying https://github.com/tc39/proposal-temporal/blob/main/polyfill/runtest262.mjs into this repo. Now I can see the failures, which seem to be related to the Intl format changes that caused the Node 19 Demitasse failures.

[justingrant]

PR filed in Test262 for the failures in Node 19: https://github.com/tc39/test262/pull/3750

I'm not sure what's going on with the test262-es5 and test262-opt runs. I assume these are rollup config issues?

[justingrant]

This PR should be ready to go, but unfortunately because Test262 changes were needed, it's blocked on catching up with the last few months of normative changes from proposal-temporal, so that we can pass the last few months of Test262 changes. When we've caught up with the normative changes then I can rebase this PR and it should be good to merge.

Caveat: there may be a problem with our ES5 and "opt" runs of Test262 with the optimized test runner from proposal-temporal. I may create a separate PR for that runner change so we can try to solve it in parallel.

I'll change to a draft PR in the meantime.

[justingrant]

@12wrigja All tests now pass because I just added a new expected-failures file to include the 897 (!!!) Test262 tests that fail because of commits that we have yet to migrate over from proposal-temporal. Seems like we have two options:

1. Merge this PR as-is, and as we migrate commits from proposal-temporal, then remove lines from the new expected-failures file.
2. Wait until proposal-temporal commits are migrated, then merge this PR (minus the new expected-failures file).

Which would you prefer? Personally I like (1) because it's one less rebase to worry about in the future, but I'm flexible either way.

[12wrigja]

Let's proceed with 1 - I don't want this work to bit-rot, and my understanding is the 262 runner will flag tests that are supposed to fail but pass and we can remove the relevant lines then.