generate-spec-types.py --check fails on main, and --apply deletes the SIGMA and DAX descriptions

[djwaldo]

scripts/generate-spec-types.py --check fails on main, and the fix it recommends deletes documentation.

Reproduce

$ git checkout main && python3 scripts/generate-spec-types.py --check
Warning: no description for 'dialects' value 'SIGMA'
Warning: no description for 'dialects' value 'DAX'
-  - "SIGMA"                 # Sigma Computing spreadsheet-style formula language
+  - "SIGMA"
-  - "DAX"                   # Data Analysis Expressions (Power BI / Analysis Services)
+  - "DAX"
core-spec/spec.yaml is out of sync with core-spec/ossie-schema.json. Run --apply to fix.
$ echo $?
1

Two separate problems

1. --apply is lossy here. Descriptions come from a hardcoded map inside the script
(generate-spec-types.py:58-67), not from the schema. SIGMA and DAX are in the schema
enum but absent from that map, so the generator emits them bare — and following the script's
own advice strips the two comments spec.yaml currently has. The repair path makes the file
worse than leaving it alone.

So the script is a second source of truth: adding a dialect to ossie-schema.json also
requires hand-editing the descriptions map, or the next --apply silently removes the
description someone wrote by hand.

2. Nothing invokes it. grep -rn generate-spec-types .github/ returns no hits, and there
is no pre-commit config or Makefile target calling it. The check exists but runs nowhere, which
is why main can sit red without anyone noticing.

Suggested fix

• Add SIGMA and DAX to the descriptions map so --check passes and --apply is
  non-destructive. (SIGMA arrived in docs(core-spec): add missing SIGMA dialect to spec.yaml enum #384, DAX alongside the Power BI work in docs(core-spec): register Power BI vendor name #413.)
• Consider making a missing description a hard error rather than a warning — as written, the
  warning is printed and generation proceeds to drop the comment, so the failure mode is
  quiet.
• Wire --check into validation-ci.yml so the gate actually gates. Without that, the same
  drift returns with the next dialect.

Happy to open a PR for any of these if useful.

Found while rebasing #364; that PR touches neither file, and I confirmed the failure on a
clean extract of main rather than from my branch.