[Task][Performance] Batch namespace snapshot cache updates and invalidation #7645

Performance baseline and program tracking: #7650.

Goal

Reduce repeated list copying, sorting and match-cache invalidation during initial synchronization and reconnect snapshots while preserving configuration semantics.

Current evidence

CommonPluginDataSubscriber.applySelectorDataSnapshot/applyRuleDataSnapshot calls per-item subscribe methods. These reach single-item upserts and invalidation. BaseDataCache already has grouped batch-refresh helpers that merge and sort once per key, so reuse existing machinery where its semantics fit. A snapshot also removes stale namespace entries; existing merge helpers alone are not a replacement for snapshot semantics.

Source at reviewed commit

Related historical work: #6796 addressed repeated sorting in cache refresh paths; this issue focuses on namespace snapshot application at the reviewed commit.

Scope and acceptance criteria

• Group selector updates per plugin and rule updates per selector; merge/sort once per affected group.
• Coalesce appropriate match-cache invalidation without retaining stale positive or negative results.
• Preserve plugin-handler callbacks, deletion semantics, ordering, namespace isolation and empty-snapshot cleanup.
• Define/test ordering against concurrent incremental updates; do not claim whole-system atomicity from per-list replacement.
• Benchmark 1k/10k+ configurations with a fixed shape, both cold initialization and reconnect, while business requests run.
• Report snapshot apply time, allocations/GC, convergence and business p99 disturbance; final configuration and routing match the reference behavior.