Add missing Auto-Compaction for cache_db

[EddieHouston]

Background

Electrs has three RocksDB databases:

• txstore_db - Stores raw transactions, blocks, confirmations
• history_db - Stores address histories and spending relationships
• cache_db - Stores aggregated stats and UTXO sets

During initial sync, auto-compaction is disabled for faster bulk writes. After sync completes, the indexer calls start_auto_compactions() to:

1. Perform one-time full compaction (blocking)
2. Mark database as compacted (marker key b"F")
3. Enable auto-compaction going forward

The Bug

In src/new_index/schema.rs around line 285-294, the update() method calls:
self.start_auto_compactions(&self.store.txstore_db); // ✓ Called
// ... processing ...
self.start_auto_compactions(&self.store.history_db); // ✓ Called
// MISSING: self.start_auto_compactions(&self.store.cache_db); // ✗ Never called

Result: cache_db never gets compacted and accumulates L0 files indefinitely.

Root Cause

Git history analysis (commit 2699560 from Jan 31, 2019):

• The cache_db was created and split out from history_db
• Looks like a forgotten start_auto_compaction call for the new cache_db

Impact

Example from production system:

  Level Files Size(MB)
  ---------------------
    L0  1372   479.17

Problems:

1. Read amplification: Every cache query must check all 1372 L0 files
2. Performance degradation: Cache lookups take longer
3. Query latency: Defeats the purpose of having a cache

Proposed Fix

Add the following to schema.rs after line 294:
self.start_auto_compactions(&self.store.cache_db);

Why This Matters

• Background cleanup: Auto-compaction will gradually merge L0 files during normal operation
• Production impact: Improves cache performance which affects all queries

NOTE:
On upgrade, this would:
1. Detect no marker key (first time)
2. Block for 10-60 seconds doing full compaction of cache_db
3. Write marker key
4. Enable auto-compaction
5. Future restarts are instant