feat(sync): auto-merge safe catalog updates

[rekram1-node]

Summary

• classify each generated sync commit before updating its PR
• enable squash auto-merge for bounded model churn with complete reasoning controls
• disable auto-merge again when a later sync becomes unsafe

Policy

• unlimited updates to existing model files, including pricing and metadata
• at most 10 created models, 10 deleted models, and 15 created plus deleted models combined
• every touched reasoning provider model must explicitly define reasoning_options, resolving reasoning through base_model

Validation

• bun test packages/core/test/auto-merge.test.ts
• bun validate

[github-actions]

Action items

• [high] [violation] packages/core/script/check-sync-auto-merge.ts:6 - Check: Created/deleted churn caps (≤10 created, ≤10 deleted, ≤15 combined) must classify real model ID add/remove events. Why: git diff --find-renames plus parseNameStatus treating R* as a single updated on the new path can turn delete+add pairs into updates when TOML bodies are similar (common for short base_model + [cost] files). Sync never renames IDs—it deletes missing models and creates new ones—so a large catalog replacement can undercount churn and auto-merge past the stated limits. Action: Drop --find-renames (plain --name-status), or count each rename as deleted(old)+created(new); update the rename unit test to match.