docs: populate benchmark tables with GCP results

[kacy]

summary

• fill in benchmark tables with real data from GCP c2-standard-8 run (2026-02-15)
• encryption overhead: 44% SET (P=16), 16% GET (P=16), 0% GET (P=1)
• vector: add qdrant column and gRPC results — ember 1,372 q/s via gRPC at 36MB RSS
• gRPC vs RESP3: RESP3 pipelined dominates (83-109k ops/s), gRPC unary at 6.3k
• pub/sub: 11k msg/s single subscriber, 27k msg/s fan-out at 100 subscribers
• update README highlights with latest numbers

what was tested

all benchmarks run on GCP c2-standard-8 (8 vCPU Intel Xeon @ 3.10GHz, 32GB RAM):

• bench/compare-redis.sh — throughput comparison
• bench/bench-memtier.sh — latency + throughput
• bench/bench-vector.sh --ember-grpc — vector with qdrant + gRPC
• bench/bench-encryption.sh — AES-256-GCM overhead
• bench/bench-grpc.sh — gRPC vs RESP3 standard ops
• bench/bench-pubsub.sh — pub/sub throughput

still pending

• memory by data type (bench-memory.sh has a grep bug — fix tracked separately)
• protobuf storage overhead (feature detection issue at runtime)
• SIFT1M recall accuracy (dataset download + longer run needed)

[kacy]

superseded by #160 which has the complete benchmark data from the final GCP run.