Configure max number of allocations per node

[dgieselaar]

Currently, when using semantic_text with an inference endpoint, Elasticsearch can scale up aggressively on ingest peaks, consuming a lot of resources. Ideally, users (me) should be able to configure a max # of allocations regardless of endpoint settings to prevent a laptop from becoming an airplane.

[elasticsearchmachine]

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

[jonathan-buttner]

To add some more context here, a Connector was being used in conjunction with semantic_text. I believe it was using the default elser inference endpoint which limits the max allocations to 32.