Eshukrun/kvbench storage bench (#1703)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* Storage-aware benchmarking with per-rank read/write operations,
storage-only and mixed RDMA+storage patterns, unified `tp_file` support,
direct-IO toggles, 4K-aligned buffers, block-splitting/sharding,
POSIX/GDS options, and concurrent async handles.
* Enhanced traffic generation CLI and outputs (hit-rate, FLOPs tuning,
memory type, warmup/iteration/isolation controls, richer metadata).

* **Reliability**
* Improved ETCD-based synchronization: configurable timeouts, more
robust rendezvous/polling, and clearer timeout diagnostics.

* **Documentation**
* Expanded KVBench README with design philosophy, diagrams, updated
examples, and quick reference.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Matan Tennenhaus <mtennenhaus@nvidia.com>
Co-authored-by: Adit Ranadive <aranadive@nvidia.com>