[Bug] `init` fails when model discovery succeeds with an empty model list

[404-Page-Found]

Description

The setup wizard treats model discovery as successful whenever fetchModels() resolves. It does not handle a successful response that contains zero models.

For OpenAI-compatible and Cohere providers, an empty model array can be returned without an HTTP/protocol error, but promptModel() immediately feeds the empty array into the interactive select() prompt. There is no manual-model fallback unless fetchModels() throws.

Location

• src/commands/init.ts:161-184 — promptModel()
• src/providers/openai-compatible.ts:99-123 — fetchModels()
• src/providers/cohere.ts — fetchModels()

Relevant code

models = await fetchModels(...);
// no empty-list check
...
const selectedModel = await select({
  message: 'Select a model:',
  options: models.map((model) => ({ value: model, label: model })),
  initialValue: existingConfig?.model,
});

The catch branch that provides Enter model name manually: runs only when fetchModels() rejects.

Steps to Reproduce

1. Configure an OpenAI-compatible or Cohere endpoint whose model-list request returns HTTP 200 with an empty array.
2. Run commit-echo init.
3. Model discovery reports success.
4. The setup wizard then attempts to display a model selection prompt with no options instead of asking for a model name.

Expected Behavior

An empty model list should fall back to the existing manual model-entry flow.

Actual Behavior

The wizard reaches an empty select() options array and cannot proceed normally through model selection.

Suggested Fix

Treat models.length === 0 as a discovery failure (or a distinct empty-result state) and invoke the manual model prompt. Add tests for HTTP 200 empty model lists for every provider that performs model discovery.

Impact

Provider endpoints that do not expose a populated model catalogue can make the primary init setup wizard unusable even though the endpoint itself is reachable and otherwise compatible.

Reviewed against current main at 6c01ad0a853501e9617a3af0a5db3b515dd8ed53.

[404-Page-Found]

Assigned to @mikemikimike. Suggested implementation: treat an empty model discovery result as a discovery failure and reuse the existing manual model-entry fallback, with regression tests for successful HTTP responses containing zero models.