feat(semantic): add output_language_override to pin summary/overview language

[0xble]

Description

Adds a new config key output_language_override that, when set, bypasses
content-based language detection in semantic summary/overview generation and
memory extraction. When empty (default), behavior is unchanged.

Content-based detection (added in #1076 to resolve #934, strengthened in
open #1521) works well when a directory's content is monolingual, but
fails on mixed-language corpora — _detect_language_from_text latches
onto minority-script content and flips the entire directory's
.overview.md / .abstract.md to a language the user does not read.
language_fallback does not help because detection runs regardless.

Symmetric impact: an English-primary deployment with some imported
Japanese/Chinese content, or a Chinese-primary deployment with embedded
English documentation, now has a single knob to pin output language.

Related Issue

Related to #934 (closed by #1076), #1067, and #1521. Complements, does
not conflict with, the source-language-following direction — override is
opt-in and defaults off.

Type of Change

• New feature (non-breaking change that adds functionality)

Changes Made

• Add output_language_override: str = "" field to OpenVikingConfig
• Add resolve_with_override(config, detect_with_fallback) as the
  canonical primitive and two thin convenience wrappers
  (resolve_output_language, resolve_output_language_from_conversation)
  in openviking/session/memory/utils/language.py
• Wire the helpers into SemanticProcessor (file summary + overview
  generation), MemoryExtractor, and SessionExtractContextProvider so
  all three language-resolution paths share one override source of truth
• Add TestOutputLanguageOverride covering override set/unset paths for
  both text and conversation resolvers (7 new tests)

Testing

• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have tested this on the following platforms:
  • Linux
  • macOS
  • Windows

$ pytest tests/storage/test_semantic_processor_language.py::TestOutputLanguageOverride \
        tests/storage/test_semantic_processor_language.py::TestLanguageDetection \
        tests/storage/test_semantic_processor_language.py::TestLanguageFlow \
        tests/storage/test_semantic_processor_language.py::TestOverviewGenerationFlow -v
======================= 25 passed, 11 warnings in 0.02s =======================

Checklist

• My code follows the project's coding style
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Example

Before (English-primary corpus with minority non-English content):

$ ov overview viking://user/<me>/memories/events
# events

このディレクトリは、システム管理、ソフトウェア開発、プロジェクト管理、および個人的な業務記録を網羅した...

## Quick Navigation

*   **システムメンテナンスとクリーンアップについて知りたい**
    *   OpenClawの削除と監査 → [mem_928970ea-...], [mem_fcd7d860-...]

Reproducible deterministically via ov reindex --regenerate --wait.

After, with output_language_override: en:

# ov.conf
semantic:
  language_fallback: en
  output_language_override: en

The detector is skipped, prompts receive Output Language: en, and the
overview is generated in English regardless of in-corpus non-English
fragments. Same mechanism works for any target language — e.g.,
output_language_override: zh-CN pins summaries to Chinese on an
English-heavy subdirectory.

Pseudocode of the change in semantic_processor.py:

# Before
fallback_language = (config.language_fallback or "en").strip() or "en"
output_language = _detect_language_from_text(content, fallback_language)

# After
output_language = resolve_output_language(content, config=config)
# where resolve_output_language honors config.output_language_override first,
# then falls back to _detect_language_from_text when override is empty.

Additional Notes

The override is strictly additive: when unset (the default), behavior is
identical to current main, so this does not interact with the open PR
#1521. They can be merged in either order.

No docs update included — happy to add a note under the semantic
summaries section of docs/en/ and docs/zh/ if maintainers prefer
that in-PR vs a follow-up.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis 🔶 934 - Partially compliant Compliant requirements: Add output_language_override config to pin summary/overview language Non-compliant requirements: Requires further human verification: 1067 - Partially compliant Compliant requirements: Add output_language_override config to bypass content-based detection Non-compliant requirements: Requires further human verification: 1521 - Partially compliant Compliant requirements: Add output_language_override config as an optional complement to source-language following Non-compliant requirements: Requires further human verification:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🏅 Score: 92

🧪 PR contains tests

🔒 No security concerns identified

✅ No TODO sections

🔀 No multiple PR themes

⚡ No major issues detected

[github-actions]

PR Code Suggestions ✨

No code suggestions found for the PR.

[qin-ctx]

Thanks for the contribution. I think we should deprecate fallback_language. If this field is set, we should log a warning.

To avoid redundant logic, I think this part can be refactored a bit. If Output Language is configured, we should always use that language. If it is not configured, we should use the detected language instead. We do not need fallback_language as an additional fallback anymore.

[0xble]

Thanks @qin-ctx — addressed in 0725fa3d. language_fallback is no longer consulted in the chain; it stays as a deprecated config field with a warning logged when a non-default value is loaded. Detection falls back to hardcoded en directly. The resolution is now strictly override -> detected -> 'en'.