feat(alibaba): add deepseek-v4-flash-0731 and glm-5.2

[cfal]

Both models are served pay-as-you-go on the international Model Studio endpoint (dashscope-intl.aliyuncs.com/compatible-mode/v1), but until now only existed under the plan providers, so callers using the API directly could not resolve them.

Pricing is the Singapore list in USD/MTok:
deepseek-v4-flash-0731 0.20 in / 0.40 out / 0.04 implicit cache
glm-5.2 1.40 in / 4.40 out / 0.28 implicit cache

reasoning_options follow the same-host siblings: Alibaba exposes reasoning_effort high|max only (low/medium map to high, xhigh to max) plus an enable_thinking toggle, and returns reasoning_content.

Sources:
https://www.alibabacloud.com/help/en/model-studio/deepseek-api https://www.alibabacloud.com/help/en/model-studio/glm https://www.alibabacloud.com/help/en/model-studio/model-pricing https://www.qwencloud.com/models/deepseek-v4-flash-0731 https://www.qwencloud.com/models/glm-5.2

[github-actions]

Action items

• [medium] [possible mistake] providers/alibaba/models/glm-5.2.toml:9 - Check: Same-host / lab baseline for reasoning_options on GLM-5.2. Why: This file authors toggle + effort high|max, but established Alibaba peers for the same model (providers/alibaba-token-plan/models/glm-5.2.toml, providers/alibaba-token-plan-cn/models/glm-5.2.toml) use effort-only high|max with no toggle, and first-party providers/zhipuai/models/glm-5.2.toml / providers/zai/models/glm-5.2.toml are also effort-only. The leading comment mixes Zhipu’s reasoning_effort none|minimal skip path with Alibaba enable_thinking, which suggests a blended control surface rather than a verified DashScope API. Action: Either drop toggle to match Alibaba same-host GLM-5.2 peers (and keep effort high|max), or keep toggle only with host-specific evidence that DashScope exposes a real enable_thinking control for glm-5.2; also fix the wire comment so it describes only the controls this host actually accepts.