Avoid running can_match in search_shards for ES|QL

[dnhatn]

We should keep the coordinator-level can_match in search_shards for ES|QL, but skip the node-level can_match, combining this step into the data node requests to reduce latency.

Since we currently don't run this step for data node requests, we should only skip it for upgraded clusters.

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[jaxxnitt]

Opened an initial PR for the node-level can_match skip logic for ES|QL:
#138014