[TRTLLM-12721][fix] Bound V2 context transfer polling

[chienchunhung]

Summary

Follow-up draft PR for V2 Python transceiver bounded polling.

This keeps the V2 change separate from:

• [TRTLLM-12721][fix] Bound disagg transfer status polling #15181: C++ disagg transfer status bounded polling
• [TRTLLM-12721][fix] Add gated disagg in-flight cancellation #15238: gated C++ in-flight cancellation and transfer-buffer poison/quarantine

Behavior

• Add an explicit blocking argument to V2 TxSession.wait_complete().
• Preserve the raw TxSession API default as blocking, so existing direct transfer tests and callers keep their previous completion-barrier behavior.
• Make KvCacheTransceiverV2.check_context_transfer_status(at_least_request_num=...) use nonblocking TxSession polling for bounded scheduler polls.
• Treat None from the nonblocking TxSession poll as "not ready yet"; the request remains queued and is polled again later.
• Preserve blockAll / at_least_request_num=None as a blocking wait-all path.

Relationship to #15181

This PR is related to #15181's bounded-polling contract, but it is intentionally V2/Python-transceiver only and is opened against main to keep the diff focused. It can be reviewed after or alongside #15181 without carrying the C++ transceiver changes in this PR.

Local validation

• PYTHONPYCACHEPREFIX=/private/tmp/trtllm-pycache python -m py_compile tensorrt_llm/_torch/disaggregation/base/transfer.py tensorrt_llm/_torch/disaggregation/transceiver.py tensorrt_llm/_torch/disaggregation/native/transfer.py
• git diff --check upstream/main..HEAD
• PATH=/opt/miniconda3/bin:$PATH PRE_COMMIT_HOME=/private/tmp/trtllm-pre-commit pre-commit run --files tensorrt_llm/_torch/disaggregation/base/transfer.py tensorrt_llm/_torch/disaggregation/transceiver.py tensorrt_llm/_torch/disaggregation/native/transfer.py

Focused pytest attempt was blocked locally by missing dependency transformers.

[chienchunhung]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #54156 [ run ] triggered by Bot. Commit: cdb81d1 Link to invocation

[tensorrt-cicd]

PR_Github #54156 [ run ] completed with state SUCCESS. Commit: cdb81d1
/LLM/main/L0_MergeRequest_PR pipeline #43239 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation