feat: add runnable LlamaIndex search and extraction example

[acedatacloud-dev]

Users can already call Ace search and extraction APIs, but the SDK has no runnable LlamaIndex workflow. Add a bilingual example that wraps the published SDK in FunctionTool, converts completed extraction results into Document, and retrieves passages with source URLs through a local keyword index.

The CLI defaults to synthetic offline responses. Live calls require explicit --live, are limited to one search and 1–3 extractions, disable retries and LLM enrichment, and preserve task/trace IDs. Pending, failed or empty extraction responses cannot enter the index. An optional official LlamaIndex MCP client command only lists the hosted Serp tool schemas and requires the user's Bearer token.

Validation:

• Published acedatacloud==2026.10.12.0, llama-index-core==0.14.25: offline workflow and 8 example tests pass.
• Hosted Serp MCP: authenticated tools/list through llama-index-tools-mcp==0.6.0 / mcp==2.3.0 returned 11 tools. No search tool or billable API executed.
• Python: 245 passed, 29 live tests skipped; Ruff lint/format pass.
• TypeScript: type check passed; 87 passed, 27 live tests skipped.
• Go: vet passed; tests passed using CGO_ENABLED=1 go test -ldflags=-linkmode=external ./... (macOS Go 1.21's default linker failed before tests).
• Provider generation: no tracked generated-source differences; 11 sync-script unittests pass; diff check passes.

The new CI workflow installs public dependencies and uses normal example-test discovery. PyPI file-CDN access failed on this machine; the tested versions were installed from a mirror. No new integration package, model/provider, billing change or official-directory submission is added. Example source/CI, package publication, official listing, paid first call, billing and attribution remain separate acceptance stages.

No additional adversarial review was requested. Merge is reserved for human review; no paid validation has been run.