Add production diagnostics

OrthoConfig emits structured tracing events for discovery attempts, selected files, skips, and failures. The library does not install a subscriber; the binary should do that once during start-up. Add the subscriber with its environment-filter support:

[dependencies]
tracing-subscriber = { version = "0.3", features = ["env-filter"] }

use ortho_config::{OrthoConfig, OrthoResult};
use serde::Deserialize;

#[derive(Deserialize, OrthoConfig)]
struct Config {
    #[ortho_config(default = 8080)]
    port: u16,
}

fn main() -> OrthoResult<()> {
    tracing_subscriber::fmt()
        .with_env_filter("ortho_config=debug")
        .with_writer(std::io::stderr)
        .try_init()
        .ok();

    let config = Config::load()?;
    println!("port={}", config.port);
    Ok(())
}

Do not log configuration values or candidate paths around these events. The crate's own diagnostics avoid path and value fields because configuration locations and values may be sensitive.

Metrics are a low-cost opt-in when the application already has a metrics recorder:

[dependencies]
ortho_config = { version = "0.9.0", features = ["metrics"] }

The feature emits bounded counters such as discovery attempts, outcomes, and failures. OrthoConfig never installs a recorder, and enabling the feature does nothing visible until the application installs one.