feat: port core SDLC graders to hud.native

[nancyjlau]

ported the core generic grader primitives from hud-sdlc-lib into hud.native so env authors can use first-party evaluators in the SDK without depending on SDLC specific code

added hud/native/graders.py with: Grade.from_subscores, Grader, BashGrader, Grader.any / Grader.all

instead of copying SDLC local types, here we reuse hud-python's existing SubScore / EvaluationResult types. Grade.from_subscores normalizes positive weights to sum to 1.0 and preserves negative weights as penalties so it fits the current hud-python evaluation semantics. no SDLC specific graders are ported here

there are tests in hud/native/tests/test_graders.py

---

Note

Medium Risk
Introduces new grading utilities that execute arbitrary shell commands via BashGrader, which can impact safety/behavior depending on how scenarios use it. Also changes telemetry argument/result serialization, which could affect what data is emitted in traces.

Overview
Adds first-party grading primitives under hud.native (Grade.from_subscores, Grader with any/all, and BashGrader) to build EvaluationResult objects from SubScores, including weight normalization, penalty handling, name de-duping, and metadata propagation.

Introduces hud.utils.serialization (json_safe_value/json_safe_dict) and switches telemetry instrumentation to use it for more robust JSON-safe span serialization; adds test coverage and new docs page wired into the docs nav.

^{Written by Cursor Bugbot for commit f93def9. This will update automatically on new commits. Configure here.}

[lorenss-m]

Surely we have something like this in the rest of the codebase we can reuse

[lorenss-m]

Small comment in the file, otherwise would also be nice to add this into the docs! Make a SDK reference file for this

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}