Reduce repeated docstring parsing during pipeline compilation #174

The one-well official 30-case fresh-server sweep spends 100.1 s in compilation and 171.2 s in non-job setup, versus 103.8 s in execution. A compile profile of cp_tutorial_advanced_segmentation_final showed 2,211 docstring source/AST parses inside the python-introspect dependency, although the AST and fallback branches both use the same parser.

python-introspect issue #3 tracks the upstream fix. A 114-callable parity check produced identical documentation results. In full runs using the upstream patch, ExampleFly compilation was 2.26 s versus 3.98 s in an intervening unmodified control, and advanced segmentation was 3.18 s versus 6.30 s. The official 30-case fresh-server sweep is in progress.

Update the OpenHCS python-introspect submodule to the validated upstream fix, preserving the compile/execution/total benchmark distinction. Link the upstream PR and report the full-sweep result before marking the bump ready.

Completed result

The full fresh-server 30-case sweep completed 30/30 after the upstream fix. Compilation sum fell 100.065 → 58.632 s, with every case faster; measured total fell 375.020 → 311.858 s. A same-code reused-server sweep also completed 30/30 with 22.607 s compilation, 93.564 s execution, and 141.944 s per-observation total (excluding once-per-sweep startup/shutdown). Raw CSVs, the native CP baseline, and native-relative figures are committed in PR #175. Execution variation is not attributed to the compile-only patch.