feat: add api-asset-client crate and /v1/assets/models API endpoint

[yujonglee]

Summary

• Adds a new api-asset-client crate that fetches model asset metadata (URL, checksum, size) from https://api.hyprnote.com/v1/assets/models
• Uses a cached OnceCell to avoid repeated network calls; exposes a resolve_model convenience function via a global default client
• Updates local-stt plugin to resolve model download URLs from the API, falling back to the hardcoded tar_url/model_url if the API call fails
• Updates local-llm plugin to resolve model download URLs from the API, with the same fallback pattern
• Updates am crate's download method to resolve URLs from the API instead of using hardcoded S3 URLs directly
• Adds strum::Display derive to local-llm::SupportedModel to enable to_string() for asset ID resolution
• Adds a new api-asset crate that serves GET /models with the full model manifest (13 models across AmModel, WhisperModel, SupportedModel)
• Wires up the endpoint in apps/api at /v1/assets/models (no auth required, as clients need this pre-authentication)

Review & Testing Checklist for Human

• Asset ID correctness: Verify that every id in crates/api-asset/src/lib.rs exactly matches the to_string() output of the corresponding model enum variant. Especially SupportedModel variants (Llama3p2_3bQ4, Gemma3_4bQ4, HyprLLM) which use default strum Display (no custom serialization)
• Data duplication: The URLs, checksums, and sizes in api-asset are hardcoded copies from am/src/model.rs, whisper-local-model/src/lib.rs, and local-llm/src/model.rs. Verify all 13 entries match their source. If any source values have changed on main since this branch, the API data is stale
• Dual ModelAsset definition: api-asset (server) and api-asset-client (client) each define their own ModelAsset struct with the same shape — confirm the field names/types match so serialization round-trips correctly
• Test end-to-end: Deploy the API, hit GET /v1/assets/models, then test model downloads for each plugin (local-stt, local-llm, am) against the live endpoint
• Fallback path: Test model downloads when the API is unreachable to confirm fallback to hardcoded URLs still works

Notes

The /v1/assets/models endpoint is unauthenticated — this is intentional since desktop clients call it before auth. The checksum field is u32 in both server and client; note that SupportedModel::model_checksum() returns u64 (current values fit in u32 but this could diverge).

Requested by: @yujonglee
Link to Devin run

[netlify]

✅ Deploy Preview for hyprnote-storybook ready!

Name	Link
🔨 Latest commit	4b16178
🔍 Latest deploy log	https://app.netlify.com/projects/hyprnote-storybook/deploys/699695459fa54a0009ba906d
😎 Deploy Preview	https://deploy-preview-4067--hyprnote-storybook.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[netlify]

✅ Deploy Preview for hyprnote ready!

Name	Link
🔨 Latest commit	4b16178
🔍 Latest deploy log	https://app.netlify.com/projects/hyprnote/deploys/6996954542939400087c1007
😎 Deploy Preview	https://deploy-preview-4067--hyprnote.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.