Fix resume — skip chapter download when content file already exists

[apetros]

Re-running lncrawl crawl --noin --all -f epub <URL> on a half-finished novel re-downloaded every single already-saved chapter instead of resuming. With 1016/1734 chapters on disk and a long crawl interrupted by Cloudflare, I was watching it start over from zero every time.

The skip condition in CrawlerService.fetch_chapter looks fine at first glance:

if (
    not refresh
    and chapter.is_available
    and chapter.extra.get("crawler_version") == crawler.version
):
    return chapter   # skip

…but crawler_version is a declared parameter on CrawlerChapter (lncrawl/core/models.py:123), and _ModelBox.get_extras() returns only kwargs beyond declared params. So when fetch_chapter does chapter.extra.update(model.get_extras()) at save time, crawler_version is filtered out — it's never written to the JSON extra blob. chapter.extra.get("crawler_version") is None forever, never equals the int crawler.version, and the skip never fires.

Verified empirically on the test DB: every completed chapter has extra = {}. Nothing was being written.

There's a deeper hole behind this — crawler.version is set to int(file.st_mtime) of the source crawler file (lncrawl/services/sources/helper.py:152), which changes on git checkout, OneDrive sync, an editor save, etc. Even if the persistence were fixed, the comparison would be fragile. Storing the version in a proper DAO column is the long-term answer, but that's a migration and out of scope for "make resume work."

This PR keeps the change minimal: trust the on-disk file. If the chapter file exists, skip the download. Anyone who really wants to re-fetch already has --refresh.

if not refresh and chapter.is_available:
    return chapter

Test plan

• Run lncrawl crawl --noin --all -f <fmt> <URL> on a partially-downloaded novel. Confirm the progress bar zooms through the already-on-disk chapters (≈1000+ c/s) and only the missing ones actually hit the network.
• Verified locally: a 1734-chapter novel with 1016 on-disk chapters processed the first 1020 in under a second, then slowed to live download speed for the rest.

[dipu-bd]

if crawler version does not match that means it could fetch different content and should not use the cache. for example: a crawler can add an additional cleaner, and chapter content can be more polished with it in the next fetch.