[Inference API] Align Get/Update Inference APIs with index API pattern

[Rassyan]

Motivation

Existing Inference APIs (GET/POST _inference/<task_type>/<inference_id>) only support single-ID operations, creating friction for bulk management scenarios like credential rotation. This PR aligns their behavior with standard Index APIs by adding wildcard(*) and comma-separated list support.

Key Changes

1. Support wildcards (prod_*) and comma-separated IDs (model_a,model_b)
2. Batch operation implementation
3. Maintained backward compatibility for single-ID usage
4. Update API now returns endpoints[] array wrapping multiple ModelConfigurations

Usage Examples

# Batch GET with wildcard and explicit IDs
GET _inference/text_embedding/prod_*,staging_model

# Wildcard credential update
POST _inference/completion/*openai*  
{
  "service_settings": {
    "api_key": "${NEW_KEY}", 
    "rate_limit": { "requests_per_minute": 1000 }
  }
}

Standardized Response Format:

{
  "endpoints": [
    {
      "inference_id": "openai-completion",
      "task_type": "completion",
      "service": "openai",
      "service_settings": {
        "model_id": "gpt-3.5-turbo",
        "rate_limit": { "requests_per_minute": 1000 }
      }
    },
    ...  // Additional endpoints
  ]
}

Testing

• Verified cluster rolling upgrades

[elasticsearchmachine]

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

[Rassyan]

Hi @maxhniebergall, I noticed you contributed most of the related code. Would you mind taking a quick look at this PR when you have a moment? Any feedback would be greatly appreciated!