Make save_large_file robust to transient errors

[Sriram-Nambiar]

Replace wget with streaming requests and Range-based resume in save_large_file, fixing #348.

wget treats transient HTTP errors (502/503/504/429) and DNS failures as fatal, and its -t counter is a budget for the entire download rather than consecutive failures without progress.

This matches the solution settled in openzim/zimwright#7:

• Resumes downloads using HTTP Range requests from the bytes on disk.
• Only counts attempts that make no progress toward the retry limit.
• Enforces explicit connect (30s) and read (60s) timeouts to avoid streaming hangs.
• Retries transient HTTP errors (429, 500, 502, 503, 504) and connection/timeout errors with backoff and jitter.
• Restarts cleanly from scratch if the server ignores Range (200 OK).

Unit tests added covering all resume, retry, and failure scenarios.

Closes #348.