Avoid making backend requests for resources that obviously are not text/html

Currently the server downloads the file just to figure out its mime type. Simply examining the file extension should work in most cases.

Inspect the proxy request path and the code that determines a resource's MIME type. Start by tracing where the server downloads a file before deciding whether it is text/html. Done means obvious non-HTML extensions avoid the backend request while text/html handling remains unchanged.

Written by the indexing model from the issue text.