Regression on handling images with non-ASCII chars in filename

[zigarn]

Last automatic build of training-continuous-integration suddenly failed: https://app.circleci.com/pipelines/github/Zenika-Training/training-continuous-integration/213/workflows/2e616255-b848-41df-8955-21a6b3ec7375/jobs/512

Loading page http://localhost:35903/slides.html ...
/app/src/cli/cli.js:187
  throw err;
  ^

TypeError [ERR_INVALID_CHAR]: Invalid character in header content ["Content-Disposition"]
    at storeHeader (node:_http_outgoing:573:5)
    at processHeader (node:_http_outgoing:568:3)
    at ServerResponse._storeHeader (node:_http_outgoing:449:11)
    at ServerResponse.writeHead (node:_http_server:417:8)
    at module.exports (/app/node_modules/serve-handler/src/index.js:754:11) {
  code: 'ERR_INVALID_CHAR'
}

After a bit of debug, found that the issue comes from header 'Content-Disposition': 'inline; filename="Processus-intégration-livraison-5c96e38fe490eab5178d..png"', so with é in path.

Looks like latest build of sensei changed behavior on handling non-ASCII chars in filename.

Example on how removing non-ASCII chars in filenames fixes the problem: https://github.com/Zenika-Training/training-continuous-integration/commit/60d97613f786bbbd079a09d0baf81599277271b2