If all target shards, excluding skipped shards, fail, we should fail the
entire query regardless of the partial_results configuration or
skip_unavailable setting. This behavior does not fully align with the
search API, where skip_unavailable ignores all failures from remote
clusters and only fails the request when all shards in the local cluster
fail. However, we believe the proposed behavior is more sensible than
the existing behavior in the search API.

Closes elastic#128994

(cherry picked from commit 8f6f763)

# Conflicts:
#	x-pack/plugin/esql/src/internalClusterTest/java/org/elasticsearch/xpack/esql/action/EsqlNodeFailureIT.java