Reuse HTTP connections for COPC ranges

[rsasaki0109]

Summary

• retain one ureq::Agent inside HttpByteSource
• share the agent connection pool and TLS state across cloned sources, parallel range workers, and request batches
• keep URL validation, query semantics, point decoding, and the default eight-range concurrency unchanged
• record same-output public Autzen measurements

Root cause

The ureq::get and ureq::head convenience functions create a fresh agent. The previous COPC byte source used them for every hierarchy and point-chunk range, discarding connection/TLS state after each request. Parallel batches therefore repeatedly paid connection setup costs.

Real-data result

Public Autzen central ROI output remained exactly 889,058 points.

• pre-change observation: 95.99 s
• shared-pool runs: 59.69 s, 37.57 s, 61.83 s
• shared-pool median: 59.69 s
• dated same-workload improvement against the single pre-change run: 1.61× / 37.8% lower elapsed time

Network conditions are variable; this is not a portable latency guarantee.

Validation

• cargo test -p spatialrust-io --features io-copc-http copc::http::tests
• cargo test -p spatialrust-io --features io-copc-http --test copc_http_local
• cargo clippy -p spatialrust-io --features io-copc-http --all-targets -- -D warnings
• cargo clippy --workspace --all-targets -- -D warnings
• RUSTFLAGS="-D warnings" cargo test --workspace --no-run
• cargo test -p spatialrust --features mvp,mvp-http --test mvp_public_copc --release -- --ignored --nocapture mvp_http_autzen_copc_bounds_smoke