perf: optimize shared-kernel-rs utilities

[Magizhan]

Problem

The shared-kernel-rs library had several performance inefficiencies and code quality issues: unnecessary string allocations in error handling, uncompiled regex patterns evaluated on every call, redundant format!() wrappers, deprecated AWS SDK usage, and missing inline hints on hot-path utility functions. Additionally, there were no performance benchmarks to catch regressions.

Solution

Performance Optimizations

• Lazy-compiled UUID regex via once_cell::sync::Lazy in middleware utils (eliminates re-compilation per request)
• &'static str for error codes — code() returns static strings instead of allocating String on every error
• #[inline] hints on hot-path functions: get_method(), get_headers(), Deref impl, error helpers
• Reduced allocations — .first() instead of .get(0), to_owned() over to_string() for &str, direct string clones instead of format wrappers
• #[serde(skip_serializing_if)] to avoid serializing empty optional fields

Code Quality

• Updated deprecated AWS SDK — aws_config::from_env() → aws_config::defaults(BehaviorVersion::latest())
• Removed unused imports (log::info, tracing::*, UnboundedReceiver, UnboundedSender)
• Proc macro improvement — measure_duration now preserves function attributes (#[allow(...)], #[cfg(...)])
• Extracted ErrorHandler<E> type alias for readability in callapi module
• Middleware refactor — extracts headers/method/path into locals before passing by reference

Benchmarks

• Added criterion benchmarks for Redis serialization patterns, serde serialization/deserialization, and middleware path normalization/URL decoding/error formatting

Testing

• cargo build passes
• cargo clippy clean
• cargo test passes
• Run benchmarks: cargo bench
• Verify downstream services compile against updated shared-kernel

Impact

• 17 files changed (4 new benchmark files, 13 modified source files)
• Risk: LOW-MEDIUM — this is a shared library; changes affect all downstream Rust services
• Key hot-path improvements: regex compilation (once vs per-request), error code allocation elimination, inline hints
• New benchmark suite enables future performance regression detection