feat: add --shards flag and improve benchmark comparisons

[kacy]

summary

• adds --shards N flag to ember-server for controlling worker thread count (defaults to available CPU cores)
• updates benchmark script to run meaningful comparisons:
  • single-threaded: ember (1 shard) vs redis — fair comparison of per-core efficiency
  • multi-threaded: ember (all cores) vs dragonfly — thread-per-core architecture comparison
  • scaling efficiency: ember multi vs ember single — shows how well sharding scales
• adds dragonfly support (auto-detected if installed)
• adds --quick flag for reduced test matrix

tested

bash bench/compare-redis.sh --ember-only --quick

=== benchmark configuration ===
cpu cores:    8
requests:     100000
clients:      50
pipeline:     16

--- single-threaded comparison (ember 1 shard vs redis) ---
SET (64B, P=16)     1,136,363
GET (64B, P=16)     1,369,863
...

--- multi-threaded comparison (ember 8 shards vs dragonfly) ---
SET (64B, P=16)     1,219,512
GET (64B, P=16)     1,282,051
...

--- scaling efficiency (ember multi-core vs single-core) ---
SET (64B, P=16)     1,219,512     1,136,363     1.1x
...

note: scaling appears limited because redis-benchmark is single-threaded. for real multi-core stress testing, would need to run multiple benchmark processes in parallel.

design notes

• --shards 0 is rejected with an error
• shard count is logged on startup: ember server starting... shards=8
• dragonfly started with --dbfilename "" to disable persistence (matches redis --save "")