Clarify Coli first-run model download

[tens1x]

I opened this PR because I ran into a confusing real-world setup flow that made it hard to tell whether the problem was installation, environment, or model download.

What happened

This was my actual setup path:

1. I installed @marswave/coli successfully with: npm install -g @marswave/coli
2. coli was available in PATH and coli --help worked as expected.
3. However, ~/.coli was still empty, so it looked like nothing had been downloaded yet.
4. TypeNo then reported: Coli model download looks incomplete:balabala......
5. The error message was also hard to read in practice: there was additional text after that message, but the overlay UI extended beyond my screen truncated the content, so I could not see the full actionable instruction.
6. At that point, it was not obvious whether:
   • coli had failed to install
   • TypeNo had failed to find coli
   • the speech model was supposed to be downloaded during install
   • or the failure was actually happening later during first transcription

What made this especially confusing is that npm install -g @marswave/coli succeeded, so the natural assumption was that the speech engine setup was already complete.

Only after digging further did it become clear that:

• installing @marswave/coli only installs the CLI
• the speech model is downloaded separately on the first actual transcription
• if that GitHub download fails or is interrupted, users can end up seeing an incomplete-download error that feels disconnected from the install step

Why this PR

The current docs and UI already contain the necessary pieces, but they do notconnect them clearly enough for a first-time user.

In particular:

• the README does not emphasize early enough that CLI install and model download are separate stages
• the troubleshooting section suggests retrying npm install, even though the failing step may actually be the first model download
• after auto-install succeeds, the app does not clearly tell the user that the first transcription will still need to download a model
• the overlay error can truncate actionable diagnostics or render them beyond the visible area

This combination makes a successful npm install feel inconsistent with the later runtime error.

Changes

• Clarify in README.md and README_CN.md that npm install -g @marswave/coli installs the CLI only.
• Document that the speech model is downloaded on the first actual transcription into ~/.coli/models/.
• Update model-download troubleshooting to recommend retrying transcription and clearing ~/.coli/models/ leftovers instead of re-running npm install.
• Show a brief post-install hint in the app that the first transcription wil still download the speech model.
• Improve timeout/incomplete-download error messages so they point users to ~/.coli/models/ and suggest proxy/TUN mode where relevant.
• Allow error text in the overlay to wrap instead of truncating critical diagnostics into a single line.

Scope

This PR only changes documentation and user-facing setup/error messaging.
It does not change the transcription pipeline itself.

---

Note

Low Risk
Low risk: changes are limited to documentation and user-facing status/error messaging plus minor overlay layout tweaks, with no changes to transcription logic or data handling.

Overview
Clarifies in README.md and README_CN.md that npm install -g @marswave/coli installs only the CLI, and that the speech model is downloaded on the first transcription into ~/.coli/models/ (including updated troubleshooting guidance to retry transcription and clear partial model files).

Updates the app’s user messaging to match this flow: shows a brief post-install hint after auto-install, improves incomplete-model/timeout diagnostics to point at ~/.coli/models/ and proxy/TUN mode, and adjusts the overlay to wrap and display longer error messages (smaller font, multi-line, wider panel).

^{Written by Cursor Bugbot for commit a5ffa3e. This will update automatically on new commits. Configure here.}