fix: shorten quest analysis batches

Sync GitHub commit 4dce03a; reduce MiniCPM quest refresh batch size to fit ZeroGPU allocations.

README.md

@@ -219,7 +219,7 @@ ADVISOR_ADAPTER_ID=build-small-hackathon/hackathon-advisor-minicpm5-lora
 ADVISOR_ADAPTER_REVISION=25de69bcde397e1bcdd852923b56a42f10222650
 ADVISOR_QUEST_ANALYZER_BACKEND=minicpm-transformers
 ADVISOR_QUEST_ADAPTER_ID=artifacts/quest-lora
-ADVISOR_QUEST_ANALYSIS_BATCH_SIZE=24
+ADVISOR_QUEST_ANALYSIS_BATCH_SIZE=8
 ADVISOR_CACHE_DIR=/data/advisor-cache
 ADVISOR_REFRESH_EMBEDDING_TIMEOUT_SECONDS=1800
 ADVISOR_EMBEDDING_MODEL_REPO=ggml-org/embeddinggemma-300m-qat-q8_0-GGUF

app.py

@@ -63,7 +63,7 @@ MAX_AUDIO_UPLOAD_BYTES = 25 * 1024 * 1024
 AUDIO_UPLOAD_SUFFIXES = {".aac", ".aif", ".aiff", ".flac", ".m4a", ".mp3", ".oga", ".ogg", ".opus", ".wav", ".webm"}
 DEFAULT_HF_ORG = "build-small-hackathon"
 DEFAULT_REFRESH_EMBEDDING_TIMEOUT_SECONDS = 1800
-DEFAULT_QUEST_ANALYSIS_BATCH_SIZE = 24
+DEFAULT_QUEST_ANALYSIS_BATCH_SIZE = 8
 REFRESH_SUBPROCESS_LOG_TAIL_LINES = 80
 REFRESH_STAGE_LABELS = {
     "crawling": "Fetching public Spaces",