Add a new AlarmOutcome and exceptionDescription for AlarmResult

[jqmmes]

We want to propagate the exception details via runAlarm. But because alarms use a ForkedPromise we can't copy the kj::String.
We set up a new alarmOutcome, that's copyable, and updated the AlarmResult to only include the exceptionDescription in the path where it's not copied

[ask-bonk]

Now let me post the review. Looking at the analysis:

1. The AlarmOutcome / AlarmResult split is the right approach to solve the ForkedPromise copyability issue - clean design.
2. The capnp schema change (adding field @3) is wire-compatible.
3. The kj::mv(result) fix in the .then() at line 548 of global-scope.c++ correctly avoids copying the kj::String.
4. The asOutcome() method is clean.

The one real issue I see is the EXCEEDED_CPU path not propagating errorDescription. Let me verify the line numbers more carefully using the diff:

Looking at the diff for global-scope.c++, the EXCEEDED_CPU path at line 482-483 is NOT part of the diff (it's unchanged code). The changed lines are:

• Line 541-544: Added .errorDescription = kj::str(description)
• Line 548: Changed to kj::mv(result)
• Line 578-581: Added .errorDescription = kj::str(e.getDescription())

The EXCEEDED_CPU path is existing code, and whether to add errorDescription there is arguably a separate concern. The PR is specifically about propagating exception details through the error paths that already have a description available.

After careful review, I think this PR is well-structured and correctly solves the problem. The design choice to split AlarmOutcome (copyable) from AlarmResult (non-copyable, has kj::String) is sound. The capnp schema change is wire-compatible. Memory safety is handled correctly with kj::mv. No actionable issues found.

LGTM

github run

[anna-cf]

Why not embed AlarmOutcome instead of duplicating fields? Is it to simplify assignments or another reason I don't see? (that's fine, just wanted to check)