docs: add MCP integration example for DeepSeek web search / fetch

[gzzwp]

What this PR does

Adds a step-by-step example showing how to connect MaxKB to DeepSeek's official
web search (POST /v1/tools/search) and web fetch (POST /v1/tools/fetch) APIs
through a standard MCP (Model Context Protocol) service.

This lets a MaxKB agent retrieve real-time web information on demand: the
knowledge base covers domain knowledge, while the MCP tool covers timeliness.

Files added

• docs/mcp/deepseek-search-mcp.md — integration guide: deploy the MCP service,
  register it in MaxKB (Tools → MCP → New), use it in applications,
  verification checklist and troubleshooting.
• examples/deepseek_search_mcp_server.py — a minimal runnable MCP service
  (~150 lines) exposing web_search and fetch_page.

Why it is useful

• Gives MaxKB users a concrete, copy-pasteable MCP integration sample, including
  the streamable_http config and API Key injection via request headers.
• Documents common pitfalls, most notably the 127.0.0.1 trap when MaxKB runs
  inside Docker (the build(deps): bump the pip group across 1 directory with 2 updates #1 cause of "connection failed" reports).
• The example passes the API Key per request, so one service instance can serve
  multiple applications and multiple Keys without server-side reconfiguration.

Verification

The example service was tested end-to-end against a mock upstream:

• GET /healthz returns {"status":"ok","service":"deepseek-search-mcp"}
• MCP handshake succeeds; tools listed as web_search, fetch_page
• web_search / fetch_page return structured JSON on success
• Invalid Key returns {"ok": false, "error": {...}} instead of raising, so the
  conversation is not interrupted

Notes

• Documentation only — no changes to application code, no database migrations.
• Happy to add an English version or relocate the files if maintainers prefer a
  different path.