docs: add troubleshooting for Codex MCP Transport closed

[Xangel0s]

Problem

Codex users can hit MCP tool failures with message: Transport closed when using Engram memory tools.

Context

This tends to happen after operational/session-boundary changes such as:

• replacing engram.exe
• editing Codex MCP config or instruction files
• force-stopping engram processes

Expected Behavior

Operators should have clear troubleshooting and prevention guidance in docs.

Proposed Solution

Document:

1. Recovery sequence for stale MCP stdio sessions
2. Preventive restart policy after binary/config updates

Scope

Documentation only.

Validation

Observed in real Windows workflow and resolved via session restart/new chat.