feat: unified in-process worker (replace Docker containers)

[TytaniumDev]

Summary

• Replace Docker containers with direct child process execution for forge simulations and in-process log condensing/GCS upload
• Add dynamic parallelism that scales based on available CPU cores and memory instead of a fixed parallelism count
• Make run_sim.sh configurable via environment variables (FORGE_PATH, DECKS_DIR, LOGS_DIR, RUN_DECKS_DIR) so it works in both standalone container and local/unified execution modes
• Add new modules: condenser.ts (log parsing/condensing), gcs-client.ts (GCS upload), types.ts (shared types)
• Add unified worker Docker setup (unified-worker/) with Dockerfile and docker-compose for containerized deployment

Test plan

• Run the worker locally with FORGE_PATH pointed at a local Forge installation and verify simulations execute
• Verify log condensing produces valid condensed.json and analyze-payload.json
• Verify GCS upload works with valid credentials
• Test dynamic parallelism calculation on machines with varying CPU/memory
• Build and run the unified-worker Docker container end-to-end
• Verify run_sim.sh still works in the standalone forge-sim container

🤖 Generated with Claude Code

[TytaniumDev]

Code review

Found 1 issue:

1. Parallel forge runs share the same RUN_DECKS_DIR (defaults to $HOME/.forge/decks/commander in run_sim.sh). Each run does `rm -rf` and copies decks there, so when parallelism > 1 they race and overwrite each other's decks; Forge may run with the wrong decks or see removed files. Fix: set RUN_DECKS_DIR per run to a unique directory (e.g. under the job dir) and pass it in the env from `runForgeSim`.

local-worker/src/worker.ts (runForgeSim env)

forge-simulation-engine/run_sim.sh (RUN_DECKS_DIR default and rm/cp) and L94-L101

Generated with Claude Code

_{- If this code review was useful, please react with a thumbs up. Otherwise, react with a thumbs down.}