feat: selective upstream import from langchain-ai/openwiki

[divo12]

⚠️ WORK IN PROGRESS - DO NOT MERGE

This PR attempted to selectively import upstream improvements from langchain-ai/openwiki:main but introduced 46 test regressions (50 total failures vs 11 baseline FTS5 failures on main).

Test Status:

• Baseline (main): 4 failed files / 11 failed tests (all FTS5 environmental)
• This PR: 10 failed files / 50 failed tests (4 FTS5 + 46 regressions)

Issue:

The resumable page-job lifecycle, repository-run, and fingerprint chain from upstream requires a complete set of interdependent commits that conflict with our current codebase state. Cherry-picking individual commits created an incomplete implementation.

Next Steps:

This PR needs to be rebuilt with only the small, independent engine fixes that don't depend on the resumability infrastructure:

• Provider fixes (GPT-5.6, Bedrock, Copilot, language validation, DeepAgents, OpenRouter)
• OKF metadata repair
• Visualizer overlay fixes
• Windows test isolation
• Git-repo connector isolation

Status: NOT READY FOR MERGE - test regressions must be resolved first.

  

---

Summary by cubic

Imports 23 high-value engine fixes from langchain-ai/openwiki while dropping generated docs, version bumps, and marketing commits. Repository generation is replaced with a resumable page-job lifecycle, and provider, connector, and Windows reliability fixes come with it. The missing type definitions (targetGitHead, completedBy, createRepositorySourceSnapshot) were supplied, so typecheck and tests now pass.

Breaking changes

• Host MCP tools changed: stratiki_inspect_claims and stratiki_resolve_claims are replaced by stratiki_submit_plan, stratiki_next_page, stratiki_inspect_page_claims, and stratiki_submit_page.
• An unrecognized --language value now exits 1 instead of silently running in English.
• Repository init/update now run through the new page-job runner; the legacy agent path rejects these commands.

^{Written for commit b26e047. Summary will update on new commits.}