hmdb_endogenous_animal.py v1.1

Why v1.1?

The original "v1.0" waited until all HMDB IDs were extracted
before starting the network crawl. On files ≫1 GB this looked like it
"hung" and wasted memory. v1.1 switches to a streaming, producer‑
consumer design: IDs are extracted from XML and fed to worker threads
on the fly, so crawling begins immediately and RAM usage stays flat.

Key Updates

• Zero‑memory blow‑up: we never store more than the executor queue
  size (≈ workers*2) IDs at once.
• Visible progress from second 1: both XML parsing and crawl speeds
  are shown via tqdm (falls back to textual counters if tqdm missing).
• Auto‑resume: identical --resume semantics, but now we also
  create a .partial checkpoint every 5 s to guard against abrupt
  power failures.
• Py≥3.7 compatible (dropped the 3.8‑only {*}tag XML shortcut).
• Graceful shutdown: Ctrl‑C or SIGTERM stops creating new tasks, but
  in‑flight requests finish and the partial TSV is flushed.