feat: stream Reginald's replies token-by-token

The 26b is slow (~30s/call); the chat now shows the answer forming instead of
freezing until it's done. The final prose streams over SSE; tool-calling turns
stay structured (no partial tokens), so streaming kicks in for the narration.

core (@commitea/core):
- chat-client.complete gains an optional onToken — when set, it requests
  stream:true and parses the OpenAI SSE stream, emitting content deltas and
  assembling streamed tool-call argument fragments into the final result.
- GiteaHttpResponse exposes the optional `body` stream (real fetch has it; stubs
  don't). agent-loop threads onToken to each completion.

app:
- model:chat forwards each delta to the renderer (event.sender.send); preload
  exposes model.onToken(cb) → unsubscribe. useChat accumulates the live stream
  into a growing bubble (with a cursor), replaced by the authoritative final
  content when the turn resolves. Unconfigured → scripted reply, unchanged.

Verified: 118 core tests green (2 streaming: SSE content deltas + tool-call
fragment assembly), desktop typecheck clean, 14 fixture e2e green. Live: a real
turn against gemma-4-26b assembles the correct answer via the streaming path
(live-reginald green) — the reply now renders token-by-token.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

packages/core/src/agent/chat-client.ts

@@ -48,8 +48,12 @@ export interface CompletionResult {
   toolCalls: ToolCall[]
 }
 
+/** Called with each streamed content delta (final-prose streaming). */
+export type OnToken = (delta: string) => void
+
 export interface ChatClient {
-  complete(messages: ChatMessage[], tools?: ToolDecl[]): Promise<CompletionResult>
+  /** When `onToken` is given, the response streams (SSE) and each content delta is emitted. */
+  complete(messages: ChatMessage[], tools?: ToolDecl[], onToken?: OnToken): Promise<CompletionResult>
 }
 
 /** Map our message shape to the OpenAI wire shape. */
@@ -85,7 +89,7 @@ export function createChatClient(config: ModelConfig, fetchImpl: FetchLike): Cha
   const url = `${config.baseUrl.replace(/\/+$/, '')}/chat/completions`
 
   return {
-    async complete(messages, tools) {
+    async complete(messages, tools, onToken) {
       const body: Record<string, unknown> = {
         model: config.model,
         messages: messages.map(toWireMessage),
@@ -95,11 +99,13 @@ export function createChatClient(config: ModelConfig, fetchImpl: FetchLike): Cha
         body.tools = tools.map(toWireTool)
         body.tool_choice = 'auto'
       }
+      const stream = !!onToken
+      if (stream) body.stream = true
       const res = await fetchImpl(url, {
         method: 'POST',
         headers: {
           'Content-Type': 'application/json',
-          Accept: 'application/json',
+          Accept: stream ? 'text/event-stream' : 'application/json',
           ...(config.apiKey ? { Authorization: `Bearer ${config.apiKey}` } : {}),
         },
         body: JSON.stringify(body),
@@ -108,6 +114,8 @@ export function createChatClient(config: ModelConfig, fetchImpl: FetchLike): Cha
         const text = await res.text().catch(() => '')
         throw new Error(`model completion failed (${res.status}): ${text.slice(0, 200)}`)
       }
+      if (stream && res.body) return readStream(res.body, onToken!)
+
       const json = (await res.json()) as { choices?: { message: RawChoiceMessage }[] }
       const msg = json.choices?.[0]?.message
       return {
@@ -121,3 +129,56 @@ export function createChatClient(config: ModelConfig, fetchImpl: FetchLike): Cha
     },
   }
 }
+
+/** Streamed tool-call delta: name arrives first, arguments accumulate across chunks. */
+interface RawToolDelta {
+  index: number
+  id?: string
+  function?: { name?: string; arguments?: string }
+}
+
+/** Parse an OpenAI SSE stream: emit content deltas via onToken, accumulate the final result. */
+async function readStream(body: ReadableStream<Uint8Array>, onToken: OnToken): Promise<CompletionResult> {
+  const reader = body.getReader()
+  const decoder = new TextDecoder()
+  let buffer = ''
+  let content = ''
+  const toolAcc: { id: string; name: string; arguments: string }[] = []
+
+  const handle = (data: string) => {
+    if (data === '[DONE]') return
+    let chunk: { choices?: { delta?: { content?: string; tool_calls?: RawToolDelta[] } }[] }
+    try {
+      chunk = JSON.parse(data)
+    } catch {
+      return
+    }
+    const delta = chunk.choices?.[0]?.delta
+    if (!delta) return
+    if (delta.content) {
+      content += delta.content
+      onToken(delta.content)
+    }
+    for (const tc of delta.tool_calls ?? []) {
+      const slot = (toolAcc[tc.index] ??= { id: '', name: '', arguments: '' })
+      if (tc.id) slot.id = tc.id
+      if (tc.function?.name) slot.name = tc.function.name
+      if (tc.function?.arguments) slot.arguments += tc.function.arguments
+    }
+  }
+
+  for (;;) {
+    const { done, value } = await reader.read()
+    if (done) break
+    buffer += decoder.decode(value, { stream: true })
+    const lines = buffer.split('\n')
+    buffer = lines.pop() ?? ''
+    for (const line of lines) {
+      const t = line.trim()
+      if (t.startsWith('data:')) handle(t.slice(5).trim())
+    }
+  }
+  if (buffer.trim().startsWith('data:')) handle(buffer.trim().slice(5).trim())
+
+  return { content, toolCalls: toolAcc.filter((t) => t.name).map((t) => ({ id: t.id, name: t.name, arguments: t.arguments })) }
+}