Optimize profiled XHTTP hot paths

[jacek4yang]

Summary

• add a sustained PID-scoped hotspot profiler and bilingual analysis
• remove request padding/path allocations and cache response padding headers
• reduce session lock/timer teardown overhead and share authentication metadata
• avoid redundant address formatting/resolution and inner Arc clones
• bump the release version to 0.1.1

Measured results

Focused same-process Criterion comparisons:

• borrowed path extraction/classification: 62.7 ns vs 103.2 ns (-39%)
• request padding validation: 149.3 ns vs 383.7 ns (-61%)
• cached response padding HeaderValue: 23.0 ns vs 118.2 ns (-81%)
• shared lock-free VLESS lookup: 73.2 ns vs 90.3 ns (-19%)

An alternating A/B window before the final outer-Arc change measured -4.6% server CPU/op and -81% transient RSS growth. The final macro rerun was intentionally discarded because unrelated host workload began consuming CPU; the report does not present it as a capacity result.

Validation

• scripts/gate.sh
• bash scripts/m7_e2e.sh
• cargo package --locked --allow-dirty
• cargo deny check
• cargo audit --no-fetch
• XHTTP TLS/H2, placement, and compatibility suites during development