[PureSaber]

Agent's frozen Lab URL points at a different commit from the coordinated research stack, so downstream installation can encounter conflicting direct Git requirements even when both Lab distributions report version 0.3.1. Align Agent with the agreed Lab commit 126e0ce1ae3cad7121baf7c311365ca4a7ad42c8.

Update only the Lab reference in project metadata, the regenerated lock, the documented dependency revision, and the existing governance-test constant. All other versions, Git refs, constraints, and lock annotations remain unchanged. Agent requests plain Lab; the new Lab rolling-research optional dependencies are not enabled. This PR changes no application behavior or model-output handling.

Validation:

• Regenerated the complete dev/LLM/build lock with the documented Python 3.10.22 / pip-tools 7.6.1 command and a selective Lab upgrade. Machine comparison confirms only the intended Lab reference changed.
• A fresh Python 3.12 environment installed the whole lock and editable Agent. pip check passes before and after installation; installed metadata and recursive active Git requirements resolve to their declared commits, including Lab 126e0ce....
• Full suite: 182 passed, coverage 85.54%, standard artifact branches 14/14; Ruff check/format pass. The first run preserved two old-reference assertion failures; updating the single explicit governance constant fixed them without weakening assertions.
• Actual public quant-assistant-eval export/baseline/score/self-test and quant-review --offline pass using synthetic inputs; original fixture input hashes are unchanged. No real model, paid data, or live trading was used.

The full downstream Pipeline closure still requires its coordinated Agent/A-share/ReportHub pins; this component PR does not by itself claim that final integration is complete.