Fix vstack shape: convert existing embeddings to 2D float32

Add resume-from-checkpoint: accumulate across preemptions

Fix ODP API: correct endpoint + API key + response parsing

Re-enable 10k checkpoints — accumulates across T4 preemptions

Switch to T4 GPU (less preemption), 6hr timeout

Add retries=3 for spot instance preemption resilience

Disable mid-scan checkpoints to prevent upload timeout cancellations

Use unique app name to avoid ConflictError on restart

Increase timeout to 4 hours for full 283k scan

Fix checkpoint: use integer division boundary detection

Fix checkpoint indentation

Add incremental checkpointing every 25k figures

Fix embed: mmap sequential zip scan, no EOCD needed

Fix USPTO URL: use ODP API instead of deprecated bulkdata.uspto.gov

Fix embed_modal: stream from HF instead of zip extraction

Add USPTO 2023-2025 extension script

Add scripts/cloud/process_full_corpus.py

Add scripts/cloud/create_hf_endpoint.py

Add Modal full-corpus processing job

Reorganize: scripts/cloud/pull_from_hub.py

Reorganize: scripts/cloud/patentsview_pipeline.py

Reorganize: scripts/cloud/embed_hf_api.py

Reorganize: scripts/cloud/embed_modal.py