Randomness is far from the most important problem with browser Javascript crypto; critics of browser crypto generally assumed that browser Javascript would get secure randomness at some point. The most important problem with this scheme is that there's no way to provide assurance that the browser JS runtime is what the user of an application expects it to be.
WebCrypto was chartered as a way to recreate the DRM schemes that Flash video plugins relied on, without needing plugins. You can read the working group notes to see that. It was never intended as, and cannot function as, a way of creating "server-proof" crypto apps.
That's correct. We only use the native runtime (ie, no shims) but obviously users have to trust that that's the case unless they want to inspect the code. If there's no working window.crypto (or for certain blacklisted platforms with partial implementations) we go straight to out-of-browser keygen. Furthermore, the user has to trust that we're not sending the private key back to us by some channel: they could do this fairly easily by inspecting the code but if they're uncomfortable it's easier to click the 'make keys on my server' option.
The last line just says, "We'll handle". Looks like it's truncated.
That said, if a user prefers to make keys on their own machine we create a custom bash / powershell script to create the necessary CSR and private key in a single paste - no clicking, no Q and A, and without installing anything.