fix: handle abort finish reason from SGLang

[vladnosiv]

Overview:

SGLang is able to throw errors from its insides through the Abort status:
https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/managers/schedule_batch.py#L157

Dynamo is currently unaware of this status and is sending it as-is, leading to error:

�[2m2026-01-22T10:26:21.018066Z�[0m �[33m WARN�[0m �[1mhttp-request�[0m: �[2mdynamo_runtime::pipeline::network::egress::addressed_router�[0m�[2m:�[0m 
Failed deserializing JSON to response �[3merr�[0m�[2m=�[0m
unknown variant `abort`, expected one of `eos`, `length`, `stop`, `error`, `cancelled`, `content_filter` at line 1 column
 40 �[3mjson_str�[0m�[2m=�[0m
{"data":{"data":{"finish_reason":"abort","token_ids":[],"completion_usage":{"prompt_tokens":4,"completion_tokens":0,"total_tokens":4,"prompt_tokens_details":null}}},"complete_final":false} 
�[2m�[3mmethod�[0m�[2m=�[0mPOST �[3muri�[0m�[2m=�[0m/v1/chat/completions �[3mversion�[0m�[2m=�[0mHTTP/1.1 �[3mx_request_id�[0m�[2m=�[0m"4926d6f4-2e32-42ae-8ca7-f1c094d493c2"�[0m

This PR handles this status correctly and passes the error message on.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

👋 Hi vladnosiv! Thank you for contributing to ai-dynamo/dynamo.

Just a reminder: The NVIDIA Test Github Validation CI runs an essential subset of the testing framework to quickly catch errors.Your PR reviewers may elect to test the changes comprehensively before approving your changes.

🚀

[vladnosiv]

cc @ishandhanani

[vladnosiv]

Fixed in #5659