fix(studio): preserve updates when scheduler fails

[zyn080302]

Summary

• Publish full Studio bundles with every locked wheel at the bundle root and root-relative SHA256-pinned requirements.
• • Keep thin bundle publication compatible with the flat full-bundle layout without duplicating runtime payloads.
• • Treat Scheduler/worker upgrade failures as degraded warnings so the Studio main Function update still proceeds.

Why

Older deployed Studio updaters stage only root-level wheels for Scheduler dependency builds. The nested wheelhouse contract therefore produced a Scheduler ZIP containing requirements and run.sh but none of the referenced wheels. VeFaaS uv then rejected the relative wheelhouse URL or could not resolve the referenced files. Scheduler failure also aborted before the main Studio Function submission.

Verification

• 213 focused release, offline-runtime, self-update, deploy-target, Release Server, and Scheduler tests passed.
• • Exact historical root-only, nested-wheelhouse, and current updater staging contracts produced 186 root wheels and byte-identical requirements.
• • Network-disabled Linux x86_64, Python 3.12, manylinux_2_28 uv resolution passed with --no-index and --require-hashes.
• • Full candidate: 191 members, 186 root wheels, no nested wheelhouse. Thin candidate: 14 members.
• • Ruff, Ruff format, Pyright, uv lock --check, pre-commit secret scan, and git diff --check passed.

Isolated cloud acceptance status

At PR creation time, the task-specific old Studio main Function is healthy. The old Scheduler scan Function completed dependency installation, Release, 100 percent traffic, Ready, and its minute Timer. The task-specific worker dependency task is still being reconciled after an ambiguous SDK response, so candidate cloud acceptance is not claimed complete yet. No customer Function, Release Server, public Studio release, or production latest pointer was changed.