feat(dag): Implement DAG compaction to bound storage growth

[xilosada]

Problem

DAG history grows unbounded:

// All deltas persisted forever in ContextDagDelta
// parent_hashes pruned at 10k but deltas remain in DB

Impact:

• Storage grows linearly with transaction count
• Old deltas never cleaned up
• Sync sends full history even when snapshot would be smaller

Current State

Checkpoints are partially supported:

// crates/node/src/delta_store.rs:1139
let checkpoint = CausalDelta::checkpoint(head_id, boundary_root_hash);

But no compaction logic exists to:

1. Determine when to checkpoint
2. Prune deltas older than checkpoint
3. Handle sync requests for pruned deltas

Proposed Solution

1. Add compaction configuration:

pub struct DagCompactionConfig {
    /// Minimum deltas before compaction eligible
    pub min_deltas_before_compact: usize,  // Default: 10000
    /// Keep recent history for delta sync
    pub retain_recent_count: usize,        // Default: 1000
    /// Compaction interval
    pub check_interval: Duration,          // Default: 1 hour
}

2. Implement compaction logic that:
   • Creates checkpoint at boundary
   • Prunes deltas older than checkpoint
   • Handles sync for pruned history (offer snapshot)

Acceptance Criteria

• Compaction runs periodically based on config
• Creates checkpoint delta at compaction boundary
• Prunes deltas older than checkpoint from DB
• Retains configurable recent history
• Sync falls back to snapshot for pruned deltas
• Compaction metrics exported

Risks & Mitigations

Risk	Mitigation
Data loss if checkpoint fails	Transactional checkpoint + prune
Slow nodes need old deltas	Keep generous retain_recent_count
Checkpoint verification	Verify state hash matches before prune

Priority: P3 (Medium Impact, High Effort)