feat: deploy docling first parser

fix: constrain llama cuda build

fix: reduce llama build memory

feat: switch to unsloth gguf runtime

feat: switch parser to qwen vl

feat: switch to docling first parser

feat: deploy paddleocr gemini parser

fix: use PaddlePaddle Docker base image (paddle GPU pre-installed)

fix: add pip timeout 600s + retries for Chinese CDN, use Python 3.12 default

fix: use Python 3.12 (Ubuntu 24.04 default), simplify Dockerfile, combine install steps

fix: remove paddle import verification during build (no CUDA available)

fix: remove GPU-dependent pre-download, use restructure_pages for cross-page tables, robust md extraction

fix: use CPU mode for model pre-download during Docker build

feat: v5.0.0 PaddleOCR-VL-1.5 + Gemini hybrid architecture

fix: increase max-model-len to 65536 for VLM image tokens, improve error logging

fix: total_mem -> total_memory, clean up debug CMD and start.sh

debug: use python3 CMD to test if output reaches HF logs

debug: inline bash -c to test if CMD even executes

fix: use full path /bin/bash entrypoint to ensure CMD executes

fix: use HF repo ID for vLLM model loading instead of local path

feat: upgrade to Qwen3-VL-30B-A3B, simplify auth, fix redirects

feat: hybrid VLM parser with Qwen3-VL-8B via vLLM (v2.0.0)

fix: single-line Python command for HF Docker builder

Initial commit: Docling Parser API