fix(logger): reduce logging overhead on request threads

[Trojan-254]

fix(logger): reduce logging overhead on request threads

Problem

Despite the logger running in a background thread, logging was introducing measurable latency on the request-handling path. Benchmarks showed a ~32% throughput drop and near 2x latency increase when logging was enabled:

	Requests/sec	Avg Latency
Without logs	114,971	719µs
With logs	78,739	1.19ms

Root Causes

• SimpleTime::get_current_timestamp() was called on the request thread per log message, performing a syscall, a year-loop calendar calculation, and a format! allocation before the message even reached the channel
• (*self.component).clone() was heap-allocating a new String on every log call despite component already being an Arc<String>
• format!() at call sites allocated unconditionally before the log level filter could discard the message
• channel() (unbounded) was used with blocking send(), which could synchronize request threads with the logger under load
• Log level was hardcoded to DEBUG/INFO via cfg!(debug_assertions) with no way to suppress logging at runtime

Changes

logger.rs

• Timestamp moved to the logger thread - SimpleTime::get_current_timestamp() is now called inside the consumer loop, completely off the request path
• Arc::clone instead of String::clone - LogMessage.component is now Arc<String>, making the clone a single atomic increment
• sync_channel(4096) replaces channel() - pre-allocated ring buffer with no growth allocations
• try_send() replaces send() - drops messages under saturation rather than blocking the request thread
• Lazy-format macros log_info!, log_debug!, log_warn!, log_error! - check the log level before calling format!(), so filtered messages produce zero allocations
• Logger::enabled() - exposes the level check cleanly for use by macros without making min_level public

config.rs

• Added log_level: Option<String> field to the Config struct
• Parsed from the [HTEAPOT] section in config.toml
• Defaults to None, falling back to the existing cfg!(debug_assertions) behaviour when not set

main.rs

• min_log now resolves from config.log_level first, with cfg!(debug_assertions) as the fallback
• All logger.xxx(format!(...)) call sites replaced with the lazy macros

Configuration

Log level can now be controlled at runtime via config.toml:

[HTEAPOT]
log_level = "none"   # "debug" | "info" | "warn" | "error" | "none"

Setting "none" suppresses all log output with zero allocation overhead on the request path.

Notes

Under extreme load, try_send will silently drop log entries when the channel buffer is full. This is intentional, log loss under saturation is preferable to request thread stalls. The 4096-entry buffer provides ample headroom for normal operation.