docs: update readmes with new benchmark sections and results

[kacy]

summary

updates both bench/README.md and README.md to reflect the expanded benchmark suite:

bench/README.md:

• added qdrant column to vector similarity and SIFT1M tables
• corrected SIFT1M note — 1M × 128 × 4B = 512MB, fits in 32GB (c2-standard-8)
• added new sections with results-pending tables:
  • gRPC vs RESP3 (sequential and pipelined RESP3 vs gRPC unary)
  • pub/sub throughput (1/10/100 subscribers, 64B/1KB, SUBSCRIBE vs PSUBSCRIBE)
  • protobuf storage overhead (PROTO.SET/GET/GETFIELD/SETFIELD vs raw)
  • memory by data type (string/hash/sorted set/vector)
• updated scripts table with bench-all, bench-grpc, bench-pubsub, bench-proto
• updated running benchmarks and cloud VM sections with new commands
• corrected command count from ~101 to ~106
• all existing caveat paragraphs preserved exactly as-is

README.md:

• added vector query performance to benchmark highlights
• added bench-vector, bench-grpc, bench-all to benchmark commands
• updated command count (101 → 106) and LOC (~22k → ~31k)

what was tested

• verified all markdown tables render correctly
• cross-referenced script names against actual files in bench/
• confirmed command count by checking Command enum in ember-protocol
• verified LOC estimate by counting .rs files excluding test modules