feat(mcp,cli): MCP outputSchema forwarding + zeph doctor diagnostic (#2931, #2930)

[bug-ops]

Summary

• research(mcp): forward MCP tool outputSchema to LLM for typed structured results #2931 — When mcp.forward_output_schema = true (default: false), Zeph appends a bounded "Expected output schema" hint to each MCP tool description sent to the LLM, enabling typed tool chaining and more accurate result parsing.
• research(cli): zeph doctor — startup configuration and connectivity diagnostic subcommand #2930 — New zeph doctor [--config <path>] [--json] subcommand runs 15 preflight checks (config, vault, LLM reachability, SQLite, Qdrant, skills, filesystem, MCP servers) and exits 0/1 based on pass/fail.
• URL-credential redaction (scheme://user:pass@host) added to scrub_content.

Changes

feat(mcp): outputSchema forwarding

• output_schema: Option<serde_json::Value> added to McpTool, ToolDef, ToolDefinition
• rmcp Tool.output_schema captured at both registration paths in client.rs
• Sanitizer extended: walks /output_schema with drop-on-injection; cross-ref scan runs on pre-drop snapshot; output_schemas_dropped counter
• New config: mcp.forward_output_schema = false, mcp.output_schema_hint_bytes = 512
• Claude + OpenAI backends append hint within combined description budget
• tool_cache_key now hashes description + output_schema

feat(cli): zeph doctor

• Command::Doctor variant in src/cli.rs
• src/commands/doctor.rs: 15 async checks, plain + JSON (schema_version: 1) output
• All detail strings through scrub_content (secrets + paths + URL creds)
• SQLite probe is read-only (magic-byte check, no file creation)
• Qdrant probe calls .health_check(), feature-gated
• MCP probes wrapped in tokio::spawn + timeout + JoinError::is_panic()

fix(core): URL-credential redaction

• URL_CREDS_REGEX in redact.rs strips scheme://user:pass@host in scrub_content

fix(clippy): pre-existing warnings

• Duration unit, redundant closures, doc_markdown, mutex across await, and ~15 other clippy pedantic warnings across zeph-a2a, zeph-acp, zeph-bench, zeph-channels, zeph-core, zeph-index, zeph-llm, zeph-mcp, zeph-sanitizer, zeph-tools, zeph-tui

Test plan

• cargo +nightly fmt --check — clean
• cargo clippy --workspace --features "desktop,ide,server,chat,pdf,scheduler" -- -D warnings — clean
• cargo nextest run --workspace --features "desktop,ide,server,chat,pdf,scheduler" --lib --bins — 8262 passed
• LLM serialization gate: feature touches MCP registration and LLM tool-list build path — live session test required before merge
• zeph doctor --config .local/config/testing.toml — validate output format and exit codes

Follow-up issues

• Double-sanitization on tool list refresh (P1-4) — architectural decision pending
• OpenAI insta snapshot tests for doctor redaction (GAP-2)
• output_schemas_dropped audit log event