Model handling is determined from model identifier, which prohibits the use of instance profile ARNs without the model name

AWS Supports application inference profiles, which contain only an identifier, not a model name.
The Bedrock-converse (and presumably other providers) determine what parameters to pass to a model based on the model name.

Where a configuration like the following works:

```
{
"modelId": "global.anthropic.claude-sonnet-4-5-123456-v1:0",
"maxTokens": 32000
}

```

The following does not:
```
{
"modelId": "arn:aws:bedrock:us-east-1:aaa23456:application-inference-profile/aaa1234",
"maxTokens": 32000
}
```

with the error:

`message: 'The model returned the following errors: `temperature` and `top_p` cannot both be specified for this model. Please use only one.',
`

Start at the Bedrock-converse handling that chooses request parameters from the modelId, then reproduce the failure with the application inference profile ARN configuration shown in the issue. Done means an identifier-only profile ARN is handled correctly without incorrectly sending conflicting temperature and top_p parameters.

Written by the indexing model from the issue text.