[Bugfix][Platform] Fix GLM47 tool-call finish backfill (#7710)

### What this PR does / why we need it?
This rebases the GLM47 tool-call parser fix onto `releases/v0.18.0`
after the MiniMax usage-accounting patch merged upstream on March 27,
2026.

It fixes OpenAI chat tool-call streaming for GLM47 by:
- draining terminal parser chunks that contain both the final argument
text and the closing `</tool_call>` suffix
- computing finish backfill from the tool argument bytes actually
emitted to the client, instead of trusting parser-internal buffered
state
- adding focused regression tests for finish backfill and terminal chunk
handling

### Does this PR introduce _any_ user-facing change?
Yes. GLM47 OpenAI-compatible streaming tool-call responses now emit
correct final chunks and argument payloads on `releases/v0.18.0`.

### How was this patch tested?
- `pytest -q tests/ut/patch/platform/test_patch_glm_tool_call_parser.py
tests/ut/patch/platform/test_patch_minimax_usage_accounting.py`
- `python -m pre_commit run --files
vllm_ascend/patch/platform/patch_glm_tool_call_parser.py
tests/ut/patch/platform/test_patch_glm_tool_call_parser.py
vllm_ascend/patch/platform/__init__.py vllm_ascend/patch/__init__.py`

---------

Signed-off-by: QwertyJack <7554089+QwertyJack@users.noreply.github.com>
Co-authored-by: QwertyJack <7554089+QwertyJack@users.noreply.github.com>

vllm_ascend/patch/__init__.py

@@ -211,6 +211,33 @@
 #       Remove this patch once the upstream MiniMax usage-accounting fix is in
 #       the runtime vLLM version used by vllm-ascend.
 #
+# ** 10. File: platform/patch_glm_tool_call_parser.py**
+# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+#   1. `vllm.entrypoints.openai.chat_completion.serving.OpenAIServingChat`
+#      `vllm.tool_parsers.glm4_moe_tool_parser.Glm4MoeModelToolParser`
+#    Why:
+#       GLM-4.7 / GLM-4.5 tool-call streaming on the release runtime still has
+#       two independent finish-path bugs:
+#       1. the parser can leave a terminal `<arg_value>... </tool_call>` chunk
+#          partially undrained, and
+#       2. finish backfill trusts the parser's internal accumulated arguments
+#          instead of the argument bytes actually sent to the client.
+#       Together these can drop a full string value or emit only a suffix like
+#       `"}` in the final SSE chunk even when non-stream output is correct.
+#    How：
+#       Monkey-patch the GLM parser to keep draining a single chunk through
+#       terminal state transitions, and monkey-patch chat streaming to track
+#       per-tool arguments actually emitted to the client before computing the
+#       finish-chunk suffix. The suffix logic still tolerates mixed JSON
+#       whitespace styles from GLM tool parsers.
+#    Related PR (if no, explain why):
+#       https://github.com/vllm-project/vllm/pull/37845
+#       https://github.com/vllm-project/vllm/pull/33218
+#    Future Plan:
+#       Remove this patch once both the GLM parser drain fix and the serving
+#       finish-backfill fix are present in the runtime vLLM version used by
+#       vllm-ascend.
+#
 # * Worker Patch:
 # ===============
 #

vllm_ascend/patch/platform/__init__.py

@@ -30,6 +30,7 @@ import vllm_ascend.patch.platform.patch_minimax_m2_config  # noqa
 import vllm_ascend.patch.platform.patch_sched_yield  # noqa
 import vllm_ascend.patch.platform.patch_torch_accelerator  # noqa
 import vllm_ascend.patch.platform.patch_minimax_usage_accounting  # noqa
+import vllm_ascend.patch.platform.patch_glm_tool_call_parser  # noqa
 
 if os.getenv("DYNAMIC_EPLB", "false").lower() in ("true", "1") or os.getenv("EXPERT_MAP_RECORD", "false") == "true":
     import vllm_ascend.patch.platform.patch_multiproc_executor  # noqa