feat(cli): add phala docs — explore live docs from the terminal (agent-friendly)

[Marvin-Cypher]

Summary

Adds a phala docs command group so users — and especially AI coding agents driving the CLI — can explore everything on docs.phala.com directly from the terminal, with zero setup. No MCP registration, no links to paste: if the agent has the phala CLI, it has the current docs.

Command	What it does
phala docs search <query>	Relevance-ranked full-text search (titles, links, excerpts)
phala docs read <page>	Read full pages by path or docs.phala.com URL
phala docs tree [path] -L <depth>	Browse the docs site as a directory tree
phala docs grep <pattern> [path]	Ripgrep across all docs content (--files for paths only)
phala docs feedback <page> <msg>	Report incorrect/outdated docs to the docs team

All subcommands support --json, require no authentication, exit non-zero with actionable hints on failure, and human output cross-links the sibling subcommands so an agent that discovers one discovers them all.

How it works

The hosted docs MCP server (https://docs.phala.com/mcp) is a stateless streamable-HTTP endpoint, so each call is a single JSON-RPC POST — implemented in a ~180-line wire-format client (cli/src/lib/docs-mcp.ts) with no MCP SDK dependency.

• search uses the server's search_phala tool
• read / tree / grep use query_docs_filesystem_phala, a virtual filesystem over the published docs that accepts cat / tree / rg commands
• read normalizes URLs → paths, strips anchors, and auto-resolves page → page.mdx → page/index.mdx
• PHALA_DOCS_MCP_URL overrides the endpoint

Because everything is served live, docs updates are visible to the CLI immediately — bundled phala help topics stay version-pinned to the installed CLI, while phala docs is the freshness channel.

Testing

• Unit tests for the filesystem-output parser, shell quoting, and path normalization (cli/src/lib/docs-mcp.test.ts)
• Full CLI suite: 416 pass / 0 fail; biome format + lint clean; tsc --noEmit clean
• Live-tested against docs.phala.com: search, tree, read (path + URL + fallback resolution), grep (--files), JSON mode, and error paths (missing page, bad tree path) all verified

Follow-ups (not in this PR)

• Mention phala docs in the top-level --help epilogue and in error remediation hints
• Document it in skills/phala-cli/SKILL.md so agents learn about it up front

🤖 Generated with Claude Code