v2.8.0 — full localization + domain-agent convergence

The package now responds in the user's language end to end.

Localization

• LLM free-text replies in the user's language — mirrors the message language even for locales with no static translations. Gated by ai-agent.ai_native.respond_in_user_language (default on). (#56)
• Full static translations for de, es, fr, pt (runtime/data_collector/lexicon/messages), with key + placeholder parity to en. (#57)
• supported_locales default is now en,ar,de,es,fr,pt so the shipped locales resolve via app/request locale out of the box. (#58)
• The runtime's directly-shown messages and the data tools' asks/errors are localized (ai-engine::runtime.responses.* / runtime.tools.*) across all 6 locales. (#58, #59)

Fixes

• Domain (ai_native) sub-agent finalizes cleanly — it now runs skill-free, so it no longer matches a global skill and gets trapped by that skill's required-final-tool guard on a read-only question. Aggregate "Total total" label deduped. (#55)

Notes

• de/es/fr/pt translations are machine-generated — a native-speaker review is recommended before production.
• Developer-facing config errors and tool success sentences (re-expressed by the model in-language) are intentionally not keyed.

Full suite green: 1909 tests.