[flthibaud]

When scraping the article page failed, the fallback read only item.Content. Plain RSS 2.0 fills that from <content:encoded>; a feed with only <description> leaves it empty, so the article was stored with no body and never fetched again.

• createArticleFromItem now uses item.Description when item.Content is empty.
• New test TestCreateArticleFromItemFallsBackToDescription: fails without the fix, passes with it.

Stacked on #54: this branch contains that PR's commit, because the test reuses the fake SQL driver it adds. Merge #54 first; this PR then shows only its own commit.

Tested with go vet and go test -race ./internal/service.

Closes #39

[flthibaud]

Review: OK to merge after #54.

• The fix is the four lines in createArticleFromItem; nothing else in the import path changes.
• Mistral's version added a second fake database/sql driver next to the one from fix: recover panics inside the article import instead of crashing the server #54. I rebased the branch onto fix: recover panics inside the article import instead of crashing the server #54 and made the test reuse recordingConnector instead, to avoid the duplicate and a merge conflict in fetcher_test.go.
• Verified the new test fails without the fix and passes with it. gofmt, vet and go test -race ./internal/service pass.