[ML] Return empty usage statistics in the event of an failure

[jonathan-buttner]

This PR refactors the inference xpack usage to return an empty response on a failure instead of a the failure. Returning a failure causes the upstream _xpack/usage to fail as well.

Ideally our get all models action would still return some models even if parsing one of them fails. We don't have that logic in place at the moment though.

Example failure:

[2025-03-13T14:48:47,408][WARN ][o.e.x.i.a.TransportInferenceUsageAction] [runTask-0] Retrieving inference usage failed with error: failed java.lang.IllegalArgumentException: failed
        at org.elasticsearch.inference@9.1.0-SNAPSHOT/org.elasticsearch.xpack.inference.action.TransportInferenceUsageAction.lambda$localClusterStateOperation$1(TransportInferenceUsageAction.java:74)
        at org.elasticsearch.server@9.1.0-SNAPSHOT/org.elasticsearch.action.ActionListener$2.onResponse(ActionListener.java:258)
        at org.elasticsearch.server@9.1.0-SNAPSHOT/org.elasticsearch.action.support.ContextPreservingActionListener.onResponse(ContextPreservingActionListener.java:33)
        at org.elasticsearch.server@9.1.0-SNAPSHOT/org.elasticsearch.action.ActionListener$3.onResponse(ActionListener.java:413)
        at org.elasticsearch.server@9.1.0-SNAPSHOT/org.elasticsearch.tasks.TaskManager$1.onResponse(TaskManager.java:203)
        at org.elasticsearch.server@9.1.0-SNAPSHOT/org.elasticsearch.tasks.TaskManager$1.onResponse(TaskManager.java:197)
        at org.elasticsearch.server@9.1.0-SNAPSHOT/org.elasticsearch.action.ActionListenerImplementations$RunBeforeActionListener.onResponse(ActionListenerImplementations.java:350)

[jonathan-buttner]

@elasticmachine test this please

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)

[elasticsearchmachine]

💚 Backport successful

Status	Branch	Result
✅	8.x