fix: 45s timeout on all LLM clients + slow contextual star beat

Timeout: All OpenAI() clients (Gemini, Cerebras, OpenRouter fallbacks)
now have timeout=45s. Previously no timeout was set — OpenRouter free
tier sometimes queues requests indefinitely, causing the tour/diagram
generation to hang at 55% for 10+ minutes. 45s is generous for any
real LLM response but catches provider hangs.

Star: contextual ✦ beat slowed to 8s cycle (was 3.6s), pink-purple
color shift at peak matches the progress bar shimmer palette.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/services/generation.py

@@ -46,6 +46,7 @@ def _openrouter_client(api_key: str):
     return OpenAI(
         api_key=api_key,
         base_url="https://openrouter.ai/api/v1",
+        timeout=45,   # OpenRouter free tier sometimes queues indefinitely — cap it
         default_headers={
             "HTTP-Referer": "http://localhost:3000",
             "X-Title": "Cartographer",
@@ -274,6 +275,7 @@ class GenerationService:
             self._client = OpenAI(
                 api_key=settings.gemini_api_key,
                 base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
+                timeout=45,
             )
             self._model  = "gemma-4-31b-it"
             print("Generation: using Gemma 4 31B (gemma-4-31b-it) via Google Gemini API")
@@ -283,6 +285,7 @@ class GenerationService:
             self._client = OpenAI(
                 api_key=settings.cerebras_api_key,
                 base_url="https://api.cerebras.ai/v1",
+                timeout=45,
             )
             # llama-3.3-70b produces dramatically better context sentences than
             # llama3.1-8b for structured tasks like "describe what this chunk does in
@@ -335,6 +338,7 @@ class GenerationService:
             self._client  = OpenAI(
                 api_key=settings.cerebras_api_key,
                 base_url="https://api.cerebras.ai/v1",
+                timeout=45,
             )
             self._model   = "llama3.3-70b"
             self.provider = "cerebras"
@@ -437,6 +441,7 @@ class GenerationService:
             self._client  = OpenAI(
                 api_key=settings.gemini_api_key,
                 base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
+                timeout=45,
             )
             self._model   = "gemma-4-31b-it"
             self.provider = "gemini"

ui/src/index.css

@@ -1739,9 +1739,18 @@ textarea:focus-visible {
 .repo-contextual {
   font-size: 9px;
   color: var(--accent-soft);
-  opacity: 0.7;
   cursor: default;
   line-height: 1;
+  /* Slow glow-beat: signals "this repo has AI context embedded" without
+     demanding attention. Opacity oscillates between dim and full while
+     text-shadow breathes the accent glow in and out. 3.6s period keeps
+     it calm — closer to a sleeping MacBook indicator than a notification. */
+  animation: contextual-beat 8s ease-in-out infinite;
+}
+@keyframes contextual-beat {
+  0%   { opacity: 0.45; color: var(--accent-soft);     text-shadow: none; }
+  50%  { opacity: 1;    color: rgba(200, 130, 255, 1); text-shadow: 0 0 8px rgba(200, 130, 255, 0.9), 0 0 16px var(--accent); }
+  100% { opacity: 0.45; color: var(--accent-soft);     text-shadow: none; }
 }
 
 .quality-tip-key {