perf: selective snapshot hydration

[fresioAS]

Description

Avoids hydrating production snapshot payloads for models and standalone audits already defined in the local project.

Previously, Context.load() hydrated every snapshot in the production environment to determine whether each node belonged to the current project or another project. The environment's compact snapshot metadata already contains the snapshot names needed to identify locally defined nodes.

This change:

• Marks locally defined production snapshot names as uncached using compact environment metadata.
• Hydrates only snapshot names absent from the local project.
• Continues hydrating missing names so deleted local nodes and cross-project nodes can be distinguished using the snapshot's project metadata.
• Preserves cross-project state merging and existing schema reconstruction behavior.
• Applies independently of project-index loading.

This primarily reduces Python-side snapshot deserialization and object hydration for projects with large production environments.

Test Plan

Live env (400 models)
plan1 40.17s <--> 33.16s 7.00s faster, 17.4%
plan2 41.41s <--> 33.36s 8.05s faster, 19.4%
plan3 38.41s <--> 32.48s 5.93s faster, 15.4%
average 40.00s <--> 33.00s 7.00s faster, 17.5%

Checklist

• I have run make style and fixed any issues
• I have added tests for my changes (if applicable)
• All existing tests pass (make fast-test)
• My commits are signed off (git commit -s) per the DCO