fix: download the file, not the page, for HF blob and Civitai page links

[mogurt]

Summary

Pasting a Hugging Face /blob/ link (the URL of the file viewer page) downloaded a ~130 KB HTML page saved under the model's file name, instead of the model. Example:

https://huggingface.co/Comfy-Org/Qwen-Image-2.1/blob/main/text_encoders/qwen3.5_9b_qwen_image_2.1_pe_i2i.int8_convrot.safetensors

URL	Response
…/blob/main/…	text/html, 127,700 bytes
…/resolve/main/…	application/octet-stream, 9,471,072,252 bytes

Civitai model page links (civitai.com/models/7808/…) had the same problem.

• ParseResult gains download_url, and the download form now hands that to aria2 instead of the pasted URL.
• Hugging Face: /blob/ and /raw/ (the latter serves the LFS pointer) are rewritten to /resolve/. Repo pages that point to no file are rejected with a hint.
• Civitai: model page links map to /api/download/models/{version}, using ?modelVersionId or, if it's missing, the model's latest version.
• Other links: if HEAD reports text/html, parsing fails with "this link opens a web page" instead of downloading it.
• Also syncs Cargo.lock's package version to 0.1.2; release-please doesn't update the lock file.

Testing

• cargo clippy --all-targets -- -D warnings, cargo test: 25 passed, with new tests for the blob/raw → resolve rewrite and Civitai model-id extraction
• npm run lint, npm run typecheck, npm test: 20 passed

End-to-end in tauri dev (separate data dir), via real IPC and the UI:

• The link above → …/resolve/main/…, 9,471,072,252 bytes, type clip.
• Civitai model page, with and without ?modelVersionId=9208 → https://civitai.com/api/download/models/9208, easynegative.safetensors, 24,655 bytes.
• HF repo page and a GitHub page → clear "page, not a file" errors.
• Real download through the form from a /blob/ link (hf-internal-testing/tiny-random-bert): aria2 received the /resolve/ URL, and the saved file is 520,212 bytes and starts with the safetensors header, not HTML.