feat(devbox): delete paused runtimes after 24 hours

[zjy365]

Summary

• delete Chat and Deploy Devboxes after they have remained paused for 24 hours
• keep Deployment Task rows, events, timeline history, runner transcripts, and deployment results permanently
• add a durable Chat Devbox lifecycle ledger and migration, and track Deploy Devbox pause timestamps
• remove the previous 30-day task purge path and the archiveAfterPauseTime Devbox contract

Why

The previous lifecycle coupled ephemeral Devbox storage to durable deployment history. Paused Devboxes could remain archived for an extended period while Deployment Task audit records were deleted after 30 days. This change separates those lifecycles: runtime storage is reclaimed promptly, while task history remains available.

Behavior and impact

• terminal Deploy Devboxes are paused, then deleted 24 hours after the confirmed pause
• Chat Devboxes are tracked before create/resume activity; confirmed pause starts their 24-hour deletion deadline
• successful deletion and 404 both converge on runtimeState=deleted
• API failures leave lifecycle state intact so the next reaper sweep retries
• no user-facing task or Devbox deletion action is added
• historical paused Deploy Devboxes are backfilled by the database migration

Validation

• bun check
• bun typecheck
• 86 focused UI tests
• helm lint charts/brain-system
• bun run db:generate (no additional schema changes)
• git diff --check

Notes

A live Devbox DELETE canary was not run because the resource API timed out and no dedicated test namespace was available. Root bun lint currently fails on unchanged packages/ui files that match main; the same ESLint failure appears in PR CI.

[aimeritething]

P1 —— 合并前应解决或拿到明确决策
连接池耗尽风险：FOR UPDATE 行锁横跨最长 10 秒的外部 Devbox 调用，且 Chat 扫描按进程 setInterval 各自运行、无跨进程租约（deploy 引擎有租约，Chat 没有）——N 个进程 × 4 并发同时钉住连接。Devbox API 一慢，无关查询就会跟着排队。建议：锁内只做状态 claim，API 调用移出事务；或给 Chat 扫描加租约。

[cla-assistant]

All committers have signed the CLA.