Yes, you're correct about this, especially in the Nostratic context, or any model that supposes that higher-order relationships between major language families do exist (there's other, comparable concepts such as "Eurasiatic", "Dene-Caucasian", or "Borean"). In my opinion, while it's certainly correct that these major families must be somehow related, I think that these macro-families are almost certainly false.
Well, "Proto-English" is perhaps the false word, but I could tell you just from the knowledge of modern English and German that something like Proto-Germanic must have existed at one point. My reconstruction would very probably be inaccurate and incomplete, but the concept by itself (that some form of Proto-Germanic existed) wouldn't be false.
Well, a proto-language must have been spoken by somebody. If you're asking wether we can identify those speakers unambigously from the archaeological context if we have no written sources, then the answer obviously has to be "no".
It depends of the time frames we are talking about. Most of the more well-established language families (such as Indo-European, Semitic, Uralic) probably date back into the early Bronze Age to Neolithic, which gives us a timeframe of some 6000 to 8000 years (the only language family that might be substantially older than this is Afroasiatic). The Nostraticists usually assume that Nostratic was spoken around the time of the Last Glacial Maximum (around 20,000 years ago), which is several times older than any of the above.
On methodology, the comparative method works well for language families (or at least, when you have multiple languages that you can compare against each other), but it breaks down when you have isolate languages because you have nothing to compare them against.