fix(provider): OpenRouterProvider::name() should return profile_id for named OpenAI-compatible profiles #691

Problem

OpenRouterProvider::name() hardcodes the return value to "openrouter" regardless of the profile. When a named OpenAI-compatible profile is configured (e.g., [providers.my-gateway] with type = "open-ai-compatible"), the provider is still identified as "openrouter" instead of using the actual profile ID.

This breaks several integration points for named profiles:

1. Auth refresh identity — the auth lifecycle cannot distinguish between different OpenAI-compatible profiles because they all report the same name.
2. Model catalog cache namespace — cached catalogs from different profiles collide under the same "openrouter" key.
3. Provider identification in logs — all named profiles log as provider=openrouter instead of their actual profile name, making debugging multi-gateway setups difficult.
4. Profile-specific behavior gates — checks like profile_id.as_deref() != Some("fpt") exist elsewhere in the same impl, confirming that profile-aware logic is expected; name() should follow the same pattern.

Affected code

File: crates/jcode-provider-openrouter-runtime/src/openrouter_provider_impl.rs
Line: ~324

Current (v0.64.2):

fn name(&self) -> &str {
    "openrouter"
}

Proposed fix

The profile_id field already exists on the struct and is used extensively throughout the same implementation for profile-specific logic. Return it when available:

fn name(&self) -> &str {
    self.profile_id.as_deref().unwrap_or("openrouter")
}

This preserves backward compatibility — when profile_id is None (vanilla OpenRouter), it still returns "openrouter". When a named profile is active, it returns the actual profile name.

Additional note

I noticed the upstream repo contains a companion test named_openai_compatible_provider_reports_profile_name (in openrouter_tests.rs as of a local hotfix on v0.32.0) that validates this exact behavior — but it was never merged upstream. That test should be included alongside this fix:

#[test]
fn named_openai_compatible_provider_reports_profile_name() {
    let _lock = ENV_LOCK.lock();
    let _key = EnvVarGuard::set("TEST_NAMED_COMPAT_KEY", "test-key");

    let profile = crate::config::NamedProviderConfig {
        base_url: "https://llm.example.com/v1".to_string(),
        api_key_env: Some("TEST_NAMED_COMPAT_KEY".to_string()),
        default_model: Some("example-model".to_string()),
        ..Default::default()
    };

    let provider = OpenRouterProvider::new_named_openai_compatible("example-compat", &profile)
        .expect("named profile should initialize");

    assert_eq!(provider.name(), "example-compat");
}

Impact

This is a one-line change with zero risk to vanilla OpenRouter users (profile_id is None → falls back to "openrouter"). It unblocks production use of multiple named OpenAI-compatible profiles behind different gateways (LiteLLM, local proxies, etc.) without identity collision.

Thank you for jcode — it has been an excellent harness for our autonomous agent pipeline (SAORI).