Port the code-fetch `Queue` and wire the syncer producer

[powerslider]

Goal

Complete the code-sync migration by porting the durable code-fetch Queue to the new p2p/proto stack and wiring it to its producer, so outstanding code requests survive restarts and the syncer's to-fetch marker handling has a real writer.

Problem

PR #5402 moved the code Syncer (consumer) into vms/evm/sync/code on the new p2p/proto client, but deferred graft/evm/sync/code/queue.go:

• The Queue is the producer-side bridge: leaf-sync workers call AddCode, which persists customrawdb to-fetch markers and forwards hashes to the syncer's channel. Its only producer is leaf-sync, which lands on the leaf branch (Add leaf-range request handler #5405), so it had no caller in feat(vms/evm/sync/code): add code-by-hash request handler #5402.
• As a result, nothing currently writes to-fetch markers. The Syncer's DeleteCodeToFetch calls (slow path and fulfill) are inert no-ops until the Queue lands, and there is no crash-recovery of outstanding code requests (init / recoverUnfetchedCodeHashes).
• graft/evm/sync/code is still live (imported by graft/evm/sync/evmstate and graft/evm/sync/engine), so it cannot be deleted until the migration completes.

Proposed Solution

1. Port queue.go into vms/evm/sync/code, adapting to the new tree. customrawdb already lives there. Drop or port a minimal replacement for graft/evm/sync/types.Finalizer.
2. Wire the leaf-sync workers to call Queue.AddCode, and pass Queue.CodeHashes() into code.NewSyncer, replacing the direct channel the feat(vms/evm/sync/code): add code-by-hash request handler #5402 integration test feeds.
3. Re-add Name() / ID() and a types.Syncer interface once a sync registry consumes them (dropped in feat(vms/evm/sync/code): add code-by-hash request handler #5402 to avoid an orphan types package).