Fix critical ModuleNotFoundError by removing absolute cloud_sre_v2 imports

Final submission: add reward curve, Colab notebook, Adversarial Designer integration, updated docs

feat: Mount interactive Gradio dashboard to root URL

feat: Hybrid OS-level + HTTP fault injection (70B Killer v2)

feat: balanced 100 SFT demos + latency injection + reward normalization

polish: professional cleanup for submission

fix: increase drain rate 50→200, generate SFT data (14 expert demos)

fix: reduce warmup queue fill from 800 to 550 for solvability

CRITICAL FIX: singleton environment to preserve state across HTTP calls

debug: add debug_info field to API response for resolution tracing

debug: add resolution check logging to trace done=False bug

fix(critical): prevent 1-step auto-resolve + missing log dirs

fix(audit): 9 interlocking bugs across all tiers

fix(critical): cascade faults now ACTUALLY degrade services

fix: command executor routing for subprocess mode + reward gaming protection + mutable default bug

fix(critical): cascade trigger now works for ALL fault types - was only checking db_lock and process_crash, now uses universal health-based detection

fix(critical): cascade episodes ending before cascade triggers - agent was solving primary fault and episode ended immediately without secondary failure

fix(critical): reset() now accepts task_id from HTTP client - was ignoring cascade/single_fault requests and always running warmup

feat: Add modern landing page for Space root URL

feat: RLVE research overhaul + 6-service architecture

fix: max_concurrent_envs=1 (single-session subprocess env)

feat: CloudSRE v2 — complete environment with adaptive sampling