chore: release v0.5.4

[github-actions]

🤖 New release

• spinsat: 0.5.3 -> 0.5.4

Changelog

0.5.4 - 2026-04-05

Added

• add WarmRandom/Cycling2 restart modes and cloud GCS instance pull
• change default integration method from Euler to Strang
• add SER convergence acceleration (--ser flag)
• activity-based clause skipping for voltage derivatives
• add Strang splitting integrator
• add Bogacki-Shampine 3(2) integrator with PI step controller
• analytical x_s update replacing numerical integration
• save per-instance Optuna trial results to Cloud SQL
• Cloud SQL as single source of truth for benchmark results
• Terraform infrastructure for distributed Optuna tuning
• Cloud Monitoring dashboard for benchmark observability
• add import blocks and fix monitoring config to match live infra
• add Terraform for persistent GCP infrastructure
• completion alert via Cloud Monitoring metric push
• GCP alerting, idle VM auto-stop, and completion notifications
• reusable monitoring.py module with custom Cloud Monitoring metrics
• GCS results durability + VM labels for cloud benchmarks
• distributed Optuna tuning on GCP spot VMs with PostgreSQL storage

Fixed

• add OOM memory guard and fix silent GCS download failure
• correct beads redirect path for crew worktree
• correct URL parsing for benchmarks DB connection on cloud VMs
• per-instance recording on cloud VMs — derive benchmarks DB URL from --db-url
• revert default integration method to Euler
• per-worker sampler seed to avoid duplicate first trials
• remove RetryFailedTrialCallback causing retry loops
• switch to MedianPruner with warmup to reduce aggressive pruning
• report per-instance progress in Optuna trials, not per-batch
• use pre-baked spinsat-optuna VM image instead of stock Debian
• spot VMs now restart on preemption instead of being deleted
• cloud Optuna worker startup — YAML patching and solver path
• extract GBD hash from filename prefix for competition data compatibility
• correct Cloud SQL user import ID format
• cloud Optuna robustness — retry loop, connection limits, idempotent DB setup
• remove extra labels from metric push (hyphen chars invalid)
• correct monitoring.py imports and VM scope for Cloud Monitoring

Other

• update CLAUDE.md and Serena memories with integration methods
• use all CPU cores for parallel instance evaluation in Optuna
• parallelize Optuna trials + fix machine type and timeout
• document hybrid DMM-CaDiCaL UNSAT detection in README
• add Serena memory for hybrid DMM-CaDiCaL UNSAT architecture
• require committing and pushing Serena memories
• commit Serena memories (cloud SQL, Optuna, benchmarking)
• add Serena MCP memory management rules and update data sections
• add competition pipeline PRD, design doc, and SAT 2025 proceedings
• add cloud_test campaign (qhid) and fix cloud_optuna for testing
• move Optuna workers into existing infra/terraform
• acknowledge Claude Code and Gas Town in README

---

This PR was generated with release-plz.