Embedded: run_embedded() missing main.rs startup parity (disk-offload recovery gate + central-accept)

[pilotspacex-byte]

Found in CodeRabbit review of PR #136 (review 4420584384), validated against merged code.

Location: src/server/embedded.rs ~221-223 and ~359-365.

Problem: run_embedded() diverges from the binary on two functional paths the PR fixed in main.rs:

1. It only restores when persistence_dir is set, so --appendonly no + disk-offload loses cold-tier recovery after restart.
2. It still enables tokio/Linux per_shard_accept, the same accept-path race/hang main.rs disabled.

Fix direction: mirror main.rs: restore when persistence_dir.is_some() || disk_offload_base.is_some(); force central-accept (per_shard_accept = false) before spawning listeners. Apply to both restore blocks.

Severity: affects in-process embedders using disk-offload and/or tokio.

[pilotspacex-byte]

Closing — fixed by commit 4828b3d7 ("recover cold tier under
appendonly=no + disable per-shard accept", explicitly fixing this issue).
run_embedded() now gates recovery on
should_run_recovery(persistence_dir, disk_offload_enabled) and sets
per_shard_accept = false, matching main.rs startup behavior.

Author: Tin Dang tindang.ht97@gmail.com