AI Gateway: Bedrock Application Inference Profile ARN fails via /compat/chat/completions

A Bedrock Application Inference Profile ARN works via AI Gateway's Bedrock native endpoint, but fails via `/compat/chat/completions`.

Works

```bash
curl -X POST \
"https://gateway.ai.cloudflare.com/v1///aws-bedrock/bedrock-runtime/us-east-2/model/arn%3Aaws%3Abedrock%3Aus-east-2%3A%3Aapplication-inference-profile%2F/invoke" \
-H "cf-aig-authorization: Bearer " \
-H "Content-Type: application/json" \
-d '{
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1024,
"messages": [{
"role": "user",
"content": [{ "type": "text", "text": "Reply only OK" }]
}]
}'
```

Response:

```json
{
"model": "claude-sonnet-...",
"type": "message",
"role": "assistant",
"content": [{ "type": "text", "text": "OK" }],
"stop_reason": "end_turn",
"usage": { "input_tokens": 13, "output_tokens": 4 }
}
```

Fails: raw ARN

```bash
curl -X POST \
"https://gateway.ai.cloudflare.com/v1///compat/chat/completions" \
-H "cf-aig-authorization: Bearer " \
-H "Content-Type: application/json" \
-d '{
"model": "aws-bedrock/arn:aws:bedrock:us-east-2::application-inference-profile/",
"messages": [{ "role": "user", "content": "Reply only OK" }],
"max_tokens": 16
}'
```

Fails: URL-encoded ARN

```json
{
"model": "aws-bedrock/arn%3Aaws%3Abedrock%3Aus-east-2%3A%3Aapplication-inference-profile%2F",
"messages": [{ "role": "user", "content": "Reply only OK" }],
"max_tokens": 16
}
```

Also tried:

```text
aws-bedrock/application-inference-profile/
aws-bedrock/
aws-bedrock/us-east-2/arn:aws:bedrock:us-east-2::application-inference-profile/
```

Question: is Bedrock Application Inference Profile ARN supported through /compat/chat/completions? If yes, what is the correct model format?

Start by reproducing the Bedrock native request and the failing /compat/chat/completions requests with raw and URL-encoded Application Inference Profile ARNs. Trace the compatibility endpoint's model parsing and compare it with the native Bedrock path; done means establishing whether these ARNs are supported and identifying the accepted model format or the missing behavior.

Written by the indexing model from the issue text.