[Linuxuser67]

TikTok direct downloads 403 because the crate's download stack cannot send what the CDN requires: its HttpHeaders has four fields (no Referer), unknown extractor headers are dropped at deserialization, and there is no cookie channel (the extraction jar is gone with the process). Upstream yt-dlp works because extraction, cookies and headers stay in one process — confirmed against the #9997 thread and tiktok.py's per-format Referer.

This moves split-part downloads onto the binary, keeping the crate for brains (selection, planning) and using the binary for transport:

• run_part_ytdlp: exact -f id + exact -o path, --newline progress folded into the existing byte-weighted row progress, shared abort channel across sequential parts, per-part timeout, log-tail failure details.
• Stale ids (rotated between dialog resolve and attempt) get one fallback-spec retry (bv* height-capped video, ba/b audio, best-single for adopted singles) before failing.
• Merge, manifest/resume, zero-byte guards and phase messages unchanged; abort still lands quiet Ok(None).

Tests: argv pinning (flags, -f/-o, trailing -- + page, cookies/UA passthrough), fallback matrix, unavailability detection, plus fake-binary integration proving spawn, progress reporting and the stale-id fallback retry end to end. 209 passing, clippy clean.