Add keepAlive flag to prevent scale-to-zero during process execution

[drappier-charles]

Summary

Adds a keepAlive flag to process execution that disables sandbox auto-hibernation (scale-to-zero) while a process runs. Uses a counter-based system via blaxel.ScaleDisable/ScaleEnable.

Key changes:

• New blaxel package (src/lib/blaxel/blaxel.go) for scale-to-zero file-based counter control with file locking
• keepAlive and timeout parameters on process request/response (HTTP + MCP)
• Timeout goroutine auto-kills keepAlive processes (default 600s; timeout=0 = infinite)
• Crash recovery: ScaleReset() on startup clears stale counter
• OpenAPI spec + LIFECYCLE.md documentation

Bug fixes (latest commits):

• StopProcess (SIGTERM path) now clears KeepAlive and calls ScaleEnable, preventing the counter from leaking when a keepAlive process is gracefully stopped
• All reads/writes of process.KeepAlive are synchronized via pm.mu (Lock/RLock) to eliminate the data race between KillProcess/StopProcess and the completion goroutine
• restartProcess reads KeepAlive/Timeout under pm.mu.RLock() to match the locking discipline

Review & Testing Checklist for Human

• Double ScaleEnable guard: Verify that when StopProcess or KillProcess clears KeepAlive=false under lock, the completion goroutine (which reads under RLock) correctly skips its own ScaleEnable call — no path should decrement the counter twice
• StopProcess error rollback: If SIGTERM fails, KeepAlive is restored under lock — confirm this doesn't leave the counter in a bad state if the process already exited between the check and the kill
• Scale file locking: writeWithLock uses LOCK_EX but GetCounter reads without a lock — verify this is acceptable (currently only used for logging)
• isScaleAvailable thread safety: The scaleAvailableChecked/scaleAvailable bools are plain globals written once at first call — confirm no goroutine can race on the initial check
• Test keepAlive end-to-end: Start a process with keepAlive=true, verify counter increments, then stop/kill and verify counter returns to 0. Test with timeout=0 (infinite) and with a short timeout that expires

Suggested test plan:

1. Deploy locally, POST a keepAlive process, confirm scale file counter > 0
2. Kill/stop the process, confirm counter returns to 0
3. Start a keepAlive process, kill sandbox-api, restart it, confirm ScaleReset resets counter to 0
4. Start two keepAlive processes, kill one — counter should be 1, not 0

Notes

• Devin Session
• Requested by: @drappier-charles

[mendral-app]

LGTM

The data race flagged in the previous review is fixed — restartProcess now reads oldProcess.KeepAlive and oldProcess.Timeout under pm.mu.RLock(), consistent with how KillProcess/StopProcess write under pm.mu.Lock(). All three race conditions from the review cycle are resolved.

_{Tag @mendral-app with feedback or questions. View session}