Add timeouts and retries to the vanilla jar download

[u9g]

DownloadContext.download fetches the vanilla jar with URL.openStream(), which has no connect or read timeout. If the CDN accepts the connection and then stalls, Paperclip waits forever: there is no exception to catch and nothing is logged after Downloading …. We hit this in CI, where a first start with an empty bundlerRepoDir hung until the job's own timeout killed it, and the only fix was to wrap Paperclip in a timeout from outside.

This opens the URL through URLConnection with a 30s connect and read timeout, and retries the download up to three times before giving up the way it does today (print the exception, exit 1).

A dropped connection mid-body is a related case: HttpURLConnection returns end-of-stream rather than throwing, so the short file only shows up at the hash check, which threw with no retry. The hash check now sits inside the same loop, so a truncated download is retried too and only the third failure throws.

Tested by hand with a small DownloadContext driver against a local HTTP server:

• headers sent, body never sent: three Read timed out after 30s each, then exit 1
• unroutable address: three connect timeouts, then exit 1
• body cut short twice, then served in full: two Hash check failed … (attempt n of 3) lines, then success
• normal download: unchanged

./gradlew build passes.