fix: convert remaining hard errors to soft diagnostics for logicalId validation [red-00]

[miliberlin]

Summary

• Non-string logicalId now coerces to string and reports a diagnostic instead of throwing a hard ValidationError
• Duplicate logicalId now tracks duplicates and reports diagnostics instead of throwing, so all errors are surfaced together
• Updates existing "should throw" tests to verify diagnostics instead

Follow-up to #1205, addressing sorccu's review comment about converting the two remaining hard error cases to soft diagnostics.

Example output

All validation errors are now reported together instead of stopping on the first one:

Parsing your project... ✅

Validating project resources... ❌
✖ Invalid property value

  The value provided for property "logicalId" is not valid.

  Reason: "boilerplate project (invalid)" contains invalid characters. Only
  A-Z, a-z, 0-9, _, -, /, #, and . are allowed.

✖ Invalid property value

  The value provided for property "logicalId" is not valid.

  Reason: A check with logicalId "duplicate-check-id" already exists.

✖ Invalid property value

  The value provided for property "logicalId" is not valid.

  Reason: A check-group with logicalId "duplicate-group-id" already exists.

✖ [ApiCheck:123] Invalid property value

  The value provided for property "logicalId" is not valid.

  Reason: Expected a string but received type "number".

✖ [ApiCheck:books api check (v1)] Invalid property value

  The value provided for property "logicalId" is not valid.

  Reason: "books api check (v1)" contains invalid characters. Only A-Z, a-z,
  0-9, _, -, /, #, and . are allowed.


✖ Unable to continue due to unresolved validation errors.

Test plan

• npx tsc --noEmit passes
• npx vitest --run src/constructs/__tests__/project.spec.ts — 7 tests pass
• Full test suite passes (excluding pre-existing playwright-check failures)
• Manual e2e test with npx checkly test confirms all diagnostics reported together

🤖 Generated with Claude Code

[miliberlin]

The first resource with a given logicalId gets stored normally in this.data, and only the second (and any subsequent) duplicate gets pushed to duplicateResources. So the diagnostic says "already exists" pointing at the duplicate, not the first occurance.

I think this is a reasonable behaviour, but if you'd prefer to flag both (so the user can see which two are clashing), we could change the approach. Wdyt? @sorccu

[miliberlin]

Fixing the tests