Load remote repo skills through session environment

[starr-openai]

Summary

• load repo-local skills through the selected session environment
• read mentioned remote SKILL.md contents through executor-backed filesystem access
• keep bundled/system skills behavior intact while isolating remote workspace cache keys

Validation

• in progress: focused remote-skills codex-core test on isolated cargo target
• transport, remote exec, workspace remap, and remote project-doc slices are already drafted in the parent PR

Stack

• base PR: Harden exec-server follow-up and workspace remap #15019

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}