[Bugfix] [Parser] Fix Qwen3 latent bug in partial params dropping values containing <

[bbrowning]

Purpose

In the new Qwen3 parser, _PARTIAL_PARAM_RE used [^<]*$ to capture in-progress parameter content. When a value contained a literal < (HTML, code, math like x<5), the regex failed to match entirely. In practice this appears to be handled gracefully by the holdback mechanism. But, it is a latent bug even though I have not been able to trigger that in the real world so far outside of the unit tests here. There could be some specific scenario of generated arguments that triggers this in a live server, so better to clean this up before users hit it.

Fix by changing [^<]*$ to (.*)$ so the capture group accepts any character including <.

Test Plan

Unit Tests

I added a couple unit tests specifically to reproduce. The test plan below runs those as well as all other parser engine, qwen3 and nemotron_v3 reasoning tests (because the nemotron parser inherits from the qwen one).

pytest -v \
  tests/parser/engine/ \
  tests/reasoning/test_qwen3_reasoning_parser.py \
  tests/reasoning/test_nemotron_v3_reasoning_parser.py \
  tests/tool_parsers/test_qwen3coder_tool_parser.py

Test Result

Unit Tests

1762 passed, 2 warnings