Moment-timezone time has method momnet.tz that takes two params:
It returns time shift (to UTC) that was at given timezone in given datetime. Ok.
The question that bothers me:
Would be grateful if someone could eliminate my doubts.
Moment-timezone uses the data from the IANA time zone database (aka the TZDB, zoneinfo, or the Olson database). Most of your questions are addressed by that data, rather than by moment-timezone itself. You'll find that other implementations (for other languages, platforms, etc.) have similar behaviors.
There is a great deal of information about how the tzdb works in the theory file in the tzdb itself, and on Wikipedia, but I'll see if I can address your specific questions:
do all the locations that currently belong to a particular timezone were in the past belonging to that zone also?
The TZDB assigns time zones based on cities (because they are less likely to changes over time than other regional boundaries). Generally, one city within a given region whose clocks have been aligned since 1970 will be chosen to represent the time in that region.
When another part of that region changes their clocks differently than the rest of the region, a new time zone is created and a new city is chosen within that region to represent the zone. We call this a "zone split". Time before the split in both zones will match (except the LMT entry), and time at the split and forward will deviate. It doesn't matter if at some point in the future the time in these regions aligns again. There are now two zones and there will continue to be - because they deviated at some point in the past.
so isn't it possible that if two locations even currently belonging to a certain timezone, in the past (even after 1970) had actually different timezones (shifts to UTC).
If there is a distinct history of timekeeping in the region, then there will be two different time zone entries. So when you say "locations", if you mean two different cities with their own time zone names in the TZDB, then by definition they don't belong to the same time zone. For example, Europe/Moscow and Europe/Volgograd are both currently in UTC+3 year-round without DST. However at the start of 1992, Moscow was UTC+3 while Volgograd was UTC+4. Their histories before then deviate even further.
On the other hand, if you are talking about a location that is not specifically referenced in the TZDB, then there is a presumption of alignment. For example, Seattle is in the US Pacific time zone, all of which is represented by America/Los_Angeles. Because there is not a unique America/Seattle, the data is representing that Seattle does not have a unique time zone history than Los Angeles.
That said - there have been a few very minor edge cases that have come up in the past where a small town that is on the boundary line between two time zones has to chose between which zone to observe. It has also happened that a small town distinctly on one side of the boundary has chosen to unofficially follow the time zone in a neighboring larger city on the other side of a boundary. These changes are sometimes mentioned on the tzdb discussion list, but are rarely recorded in the data as a distinct zone.
With these edge cases, keep in mind that the TZDB only tracks cities - not regional boundaries that may divide cities or towns. For that, you'd have to go different data source. The best one I know of is Evan Siroky's timezone-boundary-builder project.
Is it possible in principle to query tz-db for specific some kind of location, not a timezone name.
You'll have to be more specific about what you mean by "location". If you mean a latitude/longitude coordinates - then the timezone-boundary-builder data, and the projects that use them, are the route to go. They will help you resolve a tzdb identifier, which you can then use with moment-timezone or other libraries.