docs: add early benchmark numbers to readme

[kacy]

summary

adds preliminary benchmark results to the README, comparing ember, redis, and dragonfly.

benchmark results

test	ember (8 cores)	ember (1 core)	redis	dragonfly*
SET (3B, P=16)	1,176,470	602,409	1,785,714	181,488
SET (64B, P=16)	1,123,595	909,090	1,204,819	176,991
SET (1KB, P=16)	625,000	526,315	1,204,819	32,626
GET (3B, P=16)	952,381	990,099	1,136,363	23,474
GET (64B, P=16)	917,431	1,030,927	1,369,863	35,842
GET (1KB, P=16)	1,020,408	990,099	1,282,051	34,710
SET (64B, P=1)	135,135	140,845	183,486	17,611
GET (64B, P=1)	132,100	134,952	183,823	16,943

*dragonfly via Docker on macOS (not representative)

tested

• ran redis-benchmark against all four server configs
• 100k requests, 50 clients, pipeline 16

notes

• redis currently outperforms ember on this Mac laptop (M1 Pro)
• dragonfly numbers are artificially low due to Docker/macOS VM overhead
• ember's multi-core scaling is limited because redis-benchmark is single-threaded
• more rigorous Linux benchmarks with memtier_benchmark needed

also documents the new --shards flag in the configuration table.