[FEATURE] Categorize live MCP startup failures (credentials vs timeout vs config)

[hello-args]

Summary

Live MCP scans that fail to start often show a generic "MCP SERVER FAILED TO START" panel without actionable categorization — or fall through to MCPProbeError when stderr patterns don't match. multi-server MCP repositories live scans with missing SSO credentials are hard to diagnose: operators can't tell env/config issue from MCTS bug from timeout.

Infrastructure for categorization exists (startup_errors.py) but coverage is incomplete and not always reached.

---

Problem

Live scan failure modes in validation:

Category	Example cause	User needs
Missing credentials	SSO_ENV, API tokens	Set env / use --env-file (F9)
Import error	wrong Python, missing module	Fix venv / -m module (C12)
Process exit	bad command path	Fix .mcp.json launch args
Timeout	server hang	Increase timeout, check stderr
Unknown	unmatched stderr	Full stderr tail

---

Proposed solution

Parse stderr; categorize credentials/health/timeout.

Acceptance Criteria

• ifd credential failure → MISSING_CREDENTIALS category
• ModuleNotFoundError → IMPORT_ERROR with venv suggestion
• Timeout → TIMEOUT with stderr tail
• Unknown failures include stderr tail in panel
• JSON error schema documented
• Pairs with F9 --env-file suggestion in message

---