Providers: unified vocabulary, listing fix, and add-provider flow (#1508)

[santoshkumarradha]

Resolves #1508.

Summary

1. Vocabulary Unification:

   • Standardized on provider as the entity that serves models across UI, settings, commands, manual, and CLI.
   • Retired legacy terminology ("service", "model service", "default service", "connection", "custom connection", "active connection", and group head "models").
   • OpenRouter routing destination standardized on host (row host, "served by ", → hosts, tab hosts).
   • External tools standardized on account.
   • Updated and extended internal/tui3/providerword_law_test.go to enforce vocabulary across non-test strings.

2. Model Listing Fix & Background Cache Engine:

   • At launch, connected providers with empty or missing caches are fetched in the background off the UI event loop via warmEmptyProviders in app.Init().
   • ctrl+r in /model triggers concurrent refreshes for all connected providers via RefreshAllModels.
   • Providers with listing failures render explicit dim status lines (can't list models · <error> · ctrl+r retry, listing models…, or lists no models · type a model id), never an empty or missing group.

3. UI Updates (/model, Settings → Providers, /connect):

   • /model group heads show provider name, address, and model count (local 127.0.0.1:8317 · 12 models), and ends with a selectable + add a provider row.
   • Settings → Providers tab lists all providers (including OpenRouter default) with model counts and key status, offering 4 actions on enter: refresh models · rename · change key · disconnect. Host, routing, and speed guard are grouped under a routing head.
   • /connect uses group head providers and description providers and accounts · connect another.

4. Add Provider Flow & Loopback Probing:

   • Implemented + add a provider with concurrent loopback port probing (11434 Ollama, 1234 LM Studio, 8000 vLLM, 8080 llama.cpp, 8317) under "found on this machine".
   • Added live probing for OpenAI-compatible endpoints (ProbeOpenAIEndpoint) requiring keys only on 401/403.

Review Passes

• Pass 1 (Defects & Contracts): Zero I/O on draw paths verified; background fetches decoupled from UI loop; error and cancellation propagation tested; loopback port probes bounded by short timeouts.
• Pass 2 (Architecture & Software Craft): Reuses existing startModelConnect/startCustomAdd/modelEntryAnswer mint flow; clean Bubble Tea command patterns; absence and vocabulary laws passing.

—
Drafted with CodeAF · reviewed and owned by the author

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}