docs(design): BufferPool sharding benchmark plan (#1297)

[oferchen]

Summary

• Adds docs/design/buffer-pool-sharding-bench.md (114 lines) specifying the criterion harness that gates the sharded BufferPool layout from Optimize bandwidth parser ASCII scan #1295.
• Fixes the bench thread-count grid at 1 / 4 / 16 / 64 with mixed adaptive buffer sizes plus a producer-consumer arm exercising cross-thread asymmetry.
• Pins pass/fail criteria: linear scaling through 16 threads, >= 2x baseline at 64 threads, no regression below the activation gate, memory ceiling growth <= 30%.
• Calls out false-sharing on shard cache lines and hot-shard imbalance under nested rayon scopes as the two top risks, with concrete mitigations (CachePadded, per-shard hit-distribution reporting).

Test plan

• CI fmt + clippy pass on docs-only change.
• Manual review confirms the document references the correct source line ranges in crates/engine/src/local_copy/buffer_pool/.
• Implementation work in Optimize bandwidth parser ASCII scan #1295 cites this plan when the bench harness lands.