downloadservlet must rely on S3

Distributions are downloaded using an old download implementation that doesn't rely on S3.

Which leads to download standing for 14min: https://app.datadoghq.com/logs?cols=host%2Cservice%2C%40mdc.PathInfo%2C%40duration&event&from_ts=1618912800000&index=main&live=false&messageDisplay=inline&query=staging%3Aprod+-%40http.url_details.path%3A%28%22%2Fnuxeo%2Frunningstatus%22+OR+%5C%2Fcas%2A+OR+%22%2Fserver-status%22+OR+%5C%2Fnuxeo%5C%2Fjavax.faces.resource%5C%2F%2A+OR+%5C%2Fnuxeo%5C%2Forg.richfaces.resources%5C%2F%2A%29&stream_sort=%40duration%2Cdesc&to_ts=1618920000000&type=logs

Meanwhile, some errors are not properly handled as HTTP status response. Any bad request ends as a 500 error.

AC:

• Download distribution from this endpoint is done on S3;
• Missing key must end a correct HTTPS status response 404.
• If more than 1 doc matches the request key, responds 409.

Impl Idea:

• If matching blob, send a redirect to the regular Nuxeo downloads endpoint: nxblob