uicacer/hpc-as-api: v0.3.1 — Per-user Globus token submission

What's new

GlobusComputeClient.submit_streaming_inference() now accepts an optional globus_token parameter. When provided, the job is submitted under the caller's own Globus identity rather than the client's stored credentials, giving per-user SLURM-level attribution on the HPC cluster.

result = await client.submit_streaming_inference(
    messages=messages,
    model="qwen25-vl-72b",
    relay_url="wss://relay.example.com",
    globus_token=caller_globus_access_token,  # optional
)

A short-lived Executor is created per request for token-authenticated calls; the persistent executor is used for API-key callers. No breaking changes — globus_token defaults to None.

See CHANGELOG.md for full details.