Enable HTTP connection reuse (keep-alive) in `Hlquery::Client` to eliminate per-request TCP/TLS overhead

[cferrys]

Summary

Hlquery::Client currently creates an LWP::UserAgent without a connection cache, which can force new TCP/TLS handshakes per request. Add keep-alive connection reuse (and lightweight configurability) to significantly reduce latency and CPU overhead for high-throughput hlquery workloads.

Context

hlquery is designed for high-performance search; the Perl client is expected to issue many small HTTP requests (health/stats, document CRUD, search, SAM, SQL). In lib/Hlquery/Client.pm, new() constructs LWP::UserAgent->new(...) but does not configure conn_cache, so connection pooling/keep-alive reuse is not guaranteed. This is especially costly for:

• bulk indexing loops (Documents->Add/Update/Delete)
• search-heavy traffic (Search, MultiSearch, SAM->Search, SQL->Query/Search)
• TLS deployments (handshake cost dominates)

Proposed Implementation

• Update lib/Hlquery/Client.pm new() to enable connection reuse by default:
  • Add use LWP::ConnCache;
  • Configure LWP::UserAgent->new(..., keep_alive => 1, conn_cache => LWP::ConnCache->new())
• Add minimal options passthrough (backwards compatible):
  • keep_alive (default: enabled)
  • conn_cache_size (default: reasonable small value, e.g. 10–50)
• Add a short benchmark/example script under examples/ that compares repeated Health()/Search() calls with and without keep-alive (measure wall time).
• Document the new options in README.md (keep it focused on performance + production defaults).

Impact

• Lower p50/p95 latency for typical hlquery request patterns (especially under TLS).
• Reduced CPU usage and fewer ephemeral ports/socket churn on clients.
• Higher sustainable QPS from the same application instance with no API changes for existing users.