dsh-gpt-perm-strip: A DSH plugin to strip redundant sandbox permission fields from GPT series models.

The DSH community plugin `dsh-gpt-perm-strip` aims to resolve the issue where GPT series models trigger errors due to DSH's strict validation failing when carrying non-upgrade sandbox permission fields during tool calls. This plugin targets only GPT series models (including `o1`/`o3`/`o4`), removing `sandbox_permissions` and `justification` fields that are not strictly wider than the current session before tool execution. Thereby, it prevents escalation failures caused by permissions that are the same as or narrower than the current session, while preserving genuine permission upgrade operations. The plugin implements parameter cleaning by wrapping the tool execution logic, without altering the behavior of non-GPT models or modifying persisted original call records. By default, Grok and DeepSeek are ignored.