They're both basically just ergonomic issues at this point, with prefixing names and always setting lexical scoping being cultural norms.
Yes! Prefixing names as a way of emulating namespaces generally works well enough. We could get modules support the same way JS used to have them, ie. as hashes of closures. Or we could use something like the `nameless` package, which rewrites the code to add a prefix to definitions automatically. Yet nobody seems to be doing either: it's just too much of a pain with too small a gain. The situation would be different if we got first-class module support in the language, but if we're going to only emulate it anyway, then namespacing with prefixes is just the easiest way to do it that is still "good enough". Still, Racket-like modules would be great to have, can't deny that :-)
Why do you think so? My main gripe is the lack of namespaces and proper modules, but there are many languages in Top 10 on TIOBE that share this problem. The only other problem is a clunky, decades old standard library. Lack of consensus around async processing is less important, but also irritating.
Other than that, Elisp improved tremendously in the last 5 years, and it keeps improving with every release. The stdlib is getting some long overdue cleanups and sensible convenience modules are added regularly. The language features of other languages are being readily incorporated in the form of macros. The live environment along with "self documenting" code makes for a rapid dev feedback, and tools like built-in Edebug and 3rd party packages like Paredit and EROS make editing Elisp code more pleasant than most other languages.
Obviously, Elisp has many problems, it's not ideal, but it's nowhere near "terrible" in my opinion. At least not anymore.