openvino

6ed00030 - [NPUW]Optimize pipeline initialize time in online partitioner. (#34958)

[NPUW]Optimize pipeline initialize time in online partitioner. (#34958) ### Details: `getPartitioning` was running very slowly when the graph contained a large number of nodes. This PR optimized several hotspots to improve performance at scale. - `hasCycle()`: add O(1) fast-path for single-consumer producers; fix DFS to mark visited at push-time to avoid redundant stack pushes - `mergeUniques()`: replace per-group O(V) getRepGroups() scan with a pre-built rep-tag->GPtrSet index, reducing total cost from O(V^2) to O(V) - `interconnect()`: use const ref for ports_map to avoid deep copy on each call - `metaInterconnect()`: cache MetaInterconnectIO; invalidate on layer changes - `getMetaDesc()`: add cache held by snapshot, to avoid re-stringifying same node ~30x improvement for 256 chunk size: - 16K context: 200s -> 7s - 8K context: 58s ->2.5s ### Tickets: - *[EISW-209691](https://jira.devtools.intel.com/browse/EISW-209691)* ### AI Assistance: - *AI assistance used: no / yes* - *If yes, summarize how AI was used and what human validation was performed (build/tests/manual checks).* --------- Signed-off-by: intelgaoxiong <xiong.gao@intel.com>