Fix task routing bug, update yaml, improve fallback and phase instructions

Expand to 30 tickets, fix Phase 1 grader, add benchmark table

Redesign environment as proper 3-phase RL MDP

Clamp reward scores to [0.002, 0.998] instead of [0.0, 1.0]

1pm

phase 1 validation v1

v2

Initial commit with files from Support-Agent repository