Completion suggestors consume a lot of on-heap FST memory

Is your feature request related to a problem? Please describe.
OpenSearch cluster 2.5 ran into high JVMMP due to completion suggesters occupying large amount of FST on-heap store. This can also possibly grow with cat/indices or stats API loading this eagerly(FST is lazily loaded)

GET /_stats/completion

## For just one index
..
....
"autocomplete-num-01": {
      "uuid": "hs2f122231daNerOIDX-V3g",
      "primaries": {
        "completion": {
          "size_in_bytes": 3832664543
        }
      },
      "total": {
        "completion": {
          "size_in_bytes": 11497993629
        }
      }
    }

Describe the solution you'd like
Explore options to cap allocation of max FST memory. Wondering if we can move this off-heap?

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Start with the completion suggesters and the _stats/completion entry point, then trace how the lazily loaded FST is accounted for in heap memory. Compare options for capping FST allocation or moving it off-heap, and define completion when the chosen approach limits memory growth without breaking completion statistics.