fix(grader): use 0.01 and 0.99 bounds to definitively prevent edge rounding to 1.0 by OpenEnv validators

Aggressive fix: Clamp everything globally to strictly avoid 0.0 or 1.0 task scores

fix: bulletproof score clamping - add _safe_score() + NaN/Inf handling + grade endpoint safety net

fix(grader): ensure task grader outputs diverse scores strictly within (0,1)

fix: update docs and metadata to strictly (0,1) score range - no 0.0 or 1.0 references

fix(grader): ensure scores fall strictly within (0, 1) range to pass Phase 2 pipeline

Fix score ranges: clamping scores strictly within (0, 1) per validator logic

fix: restructure project for multi-mode deployment compliance