feat(sandbox): make E2B cloud isolation runnable + lease correctness fixes

[liyangbing]

What & why

The agent_daemon connector already had E2B client/provider code, but no
committed, reproducible way to build the sandbox template it needs (no
e2b.Dockerfile, no make target), so cloud isolation could not be stood up
end-to-end. Self-host compose also passed PARSAR_E2B_TEMPLATE_ID, a name
the server never reads (it reads AGENT_DAEMON_SANDBOX_TEMPLATE), silently
disabling managed sandbox mode. This PR makes cloud isolation actually
runnable and fixes several lease-lifecycle correctness bugs found while
getting it working live.

How

• Reproducible template build. infra/sandbox/e2b.Dockerfile targets
  e2b's non-BuildKit builder: no multi-stage, no persisted /tmp,
  literal-quote ARG lowering; it COPYs prebuilt linux/amd64 binaries staged
  into infra/sandbox/.build/ (gitignored) by make e2b-template. Agent
  CLI installs stay single-sourced in scripts/install-agents.sh, shared
  by both Dockerfiles.
• Auto-renew defaults ON (fail-open); opt out with
  AGENT_DAEMON_SANDBOX_AUTO_RENEW=false. Default TTL is 1h, so without
  renewal a continuously-used agent's sandbox would be reaped mid-session.
• Maintain() renew errors: transient → leave renewing for the
  crashed-claim recovery re-claim (retries, never permanently disables);
  provider-unsupported (405/501 via new e2b.IsUnsupported) → disable
  policy; gone (404) → mark killed + evict, freeing the reservation slot.
• Cost leak: persist the real sandbox id right after Create
  (SetSpawningSandboxBindingSandboxID, guarded on
  lifecycle_status='spawning') so reclaim can kill it — closes the
  crash-before-Finalize window that previously leaked a billed sandbox.
• Maintenance batch size 2 → 50; compose/.env aligned to the
  AGENT_DAEMON_SANDBOX_* names the server actually reads (kept distinct
  from the e2b-only PARSAR_E2B_* credentials).
• CreateAgentDialog sandbox auto-renew toggle (default on, sandbox mode
  only); CONTRIBUTING sandbox maintenance + two-Dockerfile rules.

Verification

• make check
• Relevant E2E target (if any): n/a
• Manual smoke: go build ./..., go test for agentdaemon / e2b /
  store / cmd/server packages, and web typecheck all pass; verified a
  live E2B acquire end-to-end through an ngrok tunnel.

Notes for reviewers

• Naming: AGENT_DAEMON_SANDBOX_* is provider-neutral (shared by the
  docker and e2b backends: backend/ttl/auto_renew/template/docker_*),
  while PARSAR_E2B_* stays e2b-only (api key, base url, host, CA). The
  compose rename aligns to what the code already read.
• Renewals are serial inside a ~30s maintenance context, so under provider
  slowness only a few of a 50-batch complete per tick (soonest-to-expire
  first, self-heals next tick). Parallelize before simultaneously-due count
  makes this the bottleneck.
• sandbox_provider.go is ~1900 lines — candidate for a follow-up split.
• An orphaned test sandbox may bill until its TTL.