fix(config): honor explicit custom model for non-DeepSeek providers (#1714)

[LeoLin990405]

Summary / 概述

EN: Fixes #1714. An explicit custom model for a non‑DeepSeek provider (e.g. deepseek --provider openai --model MiniMax-M2.7) was silently replaced by a DeepSeek/provider default, so the configured model never reached the endpoint. This PR makes an explicitly‑chosen custom model pass through verbatim for every non‑DeepSeek provider.

中文： 修复 #1714。当使用非 DeepSeek 提供方并显式指定自定义模型时（例如 deepseek --provider openai --model MiniMax-M2.7），配置的模型名会被静默替换为 DeepSeek/提供方默认值，导致请求实际发送的并不是用户指定的模型。本 PR 让任何非 DeepSeek 提供方下显式指定的自定义模型原样透传。

Root cause / 根因

EN: The CLI --model handoff only ever exports DEEPSEEK_MODEL (never the provider‑specific *_MODEL var). In apply_env_overrides, DEEPSEEK_MODEL was funneled into the root default_text_model slot — a DeepSeek‑only slot the validator rejects for non‑DeepSeek IDs. Config::default_model() then treated it as a weak, normalizable default and, for an unrecognized custom model on a non‑DeepSeek provider, fell through to a DeepSeek/provider default instead of returning the user's string. This matches the maintainer's own suspicion in the v0.8.40 audit (#1736): “explicit DEEPSEEK_MODEL or --model getting overridden later by settings.provider_models or provider‑switch state.”

中文： CLI 的 --model 透传只会导出 DEEPSEEK_MODEL（永远不会导出提供方专用的 *_MODEL 变量）。apply_env_overrides 把 DEEPSEEK_MODEL 写进了根级 default_text_model 槽位——而该槽位是 DeepSeek 专用的，校验器会拒绝非 DeepSeek 的模型 ID。随后 Config::default_model() 把它当作一个可被规范化的弱默认值；对于非 DeepSeek 提供方上无法识别的自定义模型，便回退成 DeepSeek/提供方默认值，而不是返回用户指定的字符串。这与维护者在 v0.8.40 审计（#1736）中的判断一致：“显式的 DEEPSEEK_MODEL 或 --model 之后被 settings.provider_models 或切换提供方的状态覆盖。”

Fix / 修复

Two surgical changes in crates/tui/src/config.rs, reusing the existing provider‑scoped passthrough mechanism (no new code path):

1. apply_env_overrides: for a non‑DeepSeek provider, route DEEPSEEK_MODEL into the active provider's model slot (mirroring the existing OPENAI_MODEL branch) instead of the DeepSeek‑only root slot. DeepSeek/DeepSeekCN behavior is unchanged.
2. default_model(): an explicit provider‑scoped model that is not a recognized DeepSeek alias is returned verbatim for non‑DeepSeek providers, instead of falling back to a default.

中文： 在 crates/tui/src/config.rs 中做两处精准改动，复用既有的按提供方透传机制（不新增代码路径）：① apply_env_overrides 对非 DeepSeek 提供方，将 DEEPSEEK_MODEL 写入当前提供方的模型槽位（与既有 OPENAI_MODEL 分支一致），而非 DeepSeek 专用根槽位；② default_model() 对非 DeepSeek 提供方，若显式模型不是已知 DeepSeek 别名，则原样返回，不再回退默认值。DeepSeek/DeepSeekCN 行为保持不变。

Scope / 改动边界

• Does not touch normalize_model_config, provider_passes_model_through, the validator, the existing SGLANG/VLLM/OLLAMA/NVIDIA_NIM_MODEL env branches, or any other file.
• Does not change DeepSeek/DeepSeekCN model resolution.
• 不改动上述函数与其他文件；不改变 DeepSeek 系列的模型解析。

Testing / 测试

New regression test deepseek_model_env_passes_custom_model_through_for_non_deepseek_providers (next to openai_provider_accepts_custom_model_and_base_url), covering: (a) provider=openai + DEEPSEEK_MODEL=MiniMax-M2.7, no OPENAI_MODEL, default base_url → default_model() == "MiniMax-M2.7"; (b) a non‑passthrough provider (novita) with the same custom model preserved verbatim.

cargo test -p deepseek-tui --bin deepseek-tui -- \
  deepseek_model_env_passes_custom_model_through_for_non_deepseek_providers \
  openai_provider_accepts_custom_model_and_base_url
# both pass; full config:: module: 146 passed, 0 failed

新增回归测试 deepseek_model_env_passes_custom_model_through_for_non_deepseek_providers，覆盖 OpenAI 兼容端点与一个非透传提供方两种情形；既有 openai_provider_accepts_custom_model_and_base_url 等仍全部通过。

Refs #1714, #1739, #1736.

[gemini-code-assist]

Code Review

This pull request addresses issue #1714 by ensuring that custom model names are correctly passed through for non-DeepSeek providers when specified via environment variables. The changes include logic to preserve verbatim model strings and an update to environment variable overriding to map values to the correct provider-specific configuration. A regression test was added to verify these behaviors. Feedback was provided to improve code clarity by marking logically unreachable match arms with unreachable!() in the environment override logic.

[gemini-code-assist]

The ApiProvider::Deepseek and ApiProvider::DeepseekCN match arms are logically unreachable here because they are already handled by the if branch at line 2388. While listing them ensures exhaustiveness, they should be explicitly marked as unreachable to clarify that this else block specifically handles non-DeepSeek providers and to avoid confusion for future maintainers.

            let entry = match provider {
                ApiProvider::Deepseek | ApiProvider::DeepseekCN => unreachable!("DeepSeek providers are handled in the if branch"),
                ApiProvider::NvidiaNim => &mut providers.nvidia_nim,
                ApiProvider::Openai => &mut providers.openai,
                ApiProvider::Atlascloud => &mut providers.atlascloud,
                ApiProvider::Openrouter => &mut providers.openrouter,
                ApiProvider::Novita => &mut providers.novita,
                ApiProvider::Fireworks => &mut providers.fireworks,
                ApiProvider::Sglang => &mut providers.sglang,
                ApiProvider::Vllm => &mut providers.vllm,
                ApiProvider::Ollama => &mut providers.ollama,
            };

[LeoLin990405]

Done in the latest push — the two logically-unreachable arms are collapsed to ApiProvider::Deepseek | ApiProvider::DeepseekCN => unreachable!("DeepSeek providers are handled in the if branch above (issue #1714)"), making it explicit that this else block is non-DeepSeek-only. No behavior change; existing regression test + clippy still green.

[Hmbown]

Thanks for the custom-model/provider fix. This was harvested into v0.8.40 release PR #1823 as commits b156d7d and 3adc05d, and #1823 is now CI-green. Closing as superseded by the release branch; thank you for improving third-party OpenAI-compatible provider support.