Add a strict/warning-fail CLI mode for validate-data

[cmungall]

Problem

linkml-term-validator validate-data reports label mismatches as warnings, but there is currently no CLI flag to make warnings fail the process.

In CI, this forces downstream projects to add local wrappers that parse stdout/stderr and convert WARN output into a non-zero exit code. That is brittle and repo-specific.

Requested feature

Please add a native strict mode for validate-data, for example one of:

• --strict
• --fail-on-warnings

The behavior should be:

• any warning causes a non-zero exit code
• errors continue to be non-zero as they are today
• successful validation with no warnings stays zero

Concrete repo impact

In monarch-initiative/dismech, we traced a bad HGNC ID/label pair that slipped through CI when warnings did not reliably hard-fail across validator versions. The current repo now carries a local shell wrapper solely to enforce warning-fail behavior until the CLI exposes it directly.

Native support would let downstream CI use standard tooling instead of custom wrappers.

[github-actions]

Summary

Feature request: Add --strict / --fail-on-warnings to validate-data so CI pipelines can explicitly opt into warning-as-error behavior without fragile shell wrappers.

Current state (from cli.py)

Command	--strict flag?	Default warning behavior
validate-schema	✅ Yes (lines 50–56)	Exits 0 with a ⚠️ message; only errors exit 1
validate-data	❌ No	Currently exits 1 for any result, including warnings (lines 307–339)

There is an asymmetry: validate-schema has an explicit, documented --strict contract; validate-data does not. This means:

• Downstream CI can't pin a stable --strict flag in their invocation for validate-data
• The behavior is an implementation detail that could change across versions (and apparently already has, per the dismech incident)

Concrete motivation

A bad HGNC ID/label pair slipped through monarch-initiative/dismech CI across a validator version change. That repo now carries a local shell wrapper to re-implement warning-fail behavior—exactly the kind of fragile workaround this flag would eliminate.

Implementation sketch

The pattern already exists in validate-schema. For validate-data, the change would be:

1. Add --strict parameter to validate_data() (analogous to line 50–56 in validate-schema)
2. After collecting report.results, separately count ERROR vs WARNING severity results
3. If --strict is set: exit 1 on any warning; otherwise maintain current behavior (exit 1 on errors/any result)
4. Propagate the flag through validate_all() which delegates to validate_data()

The severity check is already present (result.severity.name == "ERROR" on line 318), so the scaffolding is in place.

Notes

• The --lenient flag (existing) already goes the other direction—suppressing failures for missing term IDs. --strict would be its complement.
• For full symmetry with validate-schema, it may also be worth considering whether validate-data should default warnings to exit 0 (like validate-schema) with --strict to opt in—but that would be a breaking change and is a separate decision.

@dragon-ai-agent please implement --strict on validate-data and validate_all (when delegating to data mode) in src/linkml_term_validator/cli.py, following the existing pattern from validate-schema. Add a test in tests/test_cli.py verifying that a warning-producing validation run exits 0 without --strict and exits 1 with --strict.