Support configurable output language for LLM analyzer findings (e.g. via env var)

[geosmart]

Summary

LLMAnalyzerBase and its subclasses (semantic_security_discovery.py,semantic_quality_policy.py, semantic_developer_intent.py) hardcode ANALYZER_PROMPT as module-level Python string constants, and BASE_ANALYSIS_PROMPT in llm_analyzer_base.py has no language directive.

There's currently no way to make the free-text finding fields (message, explanation, remediation) come back in a language other than English without forking the package and editing prompt strings directly.

For non-English-speaking teams reviewing scan output (e.g. in CI comments,SARIF annotations shown to reviewers), this is a real friction point —every finding has to be manually translated or read through an LLM before it's usable internally.

Proposed change

Add a SKILLSPECTOR_OUTPUT_LANGUAGE (name open to bikeshedding) environment variable, following the same pattern as the existing
SKILLSPECTOR_MAX_LLM_CONCURRENCY:

def resolve_output_language() -> str | None:
    """Resolve the desired output language from SKILLSPECTOR_OUTPUT_LANGUAGE.

    Defaults to None (model's natural choice, currently English). When set,
    an instruction is appended to BASE_ANALYSIS_PROMPT telling the model to
    write message/explanation/remediation fields in the given language,
    while keeping rule_id values (e.g. SSD-1, SQP-3) untranslated.
    """
    return os.environ.get("SKILLSPECTOR_OUTPUT_LANGUAGE", "").strip() or None

And in BASE_ANALYSIS_PROMPT / LLMAnalyzerBase.build_prompt