You are thinking of UTC offsets as zones here, which is wrong. Yes, you can interpret an offset from the epoch in any utc offset and that's just a constant formatting operation. But interpreting a zoned datetime as an offset against a point in UTC (or UTC+/-X) is not.
You do not confidently know how far away 2025-03-01T00:00:00 America/New_York is from 1970-01-01T00:00:00+0000 until after that time. Even if you decide you're interpreting 1970-01-01T00:00:00+0000 as 1969-12-31T19:00-0500. Postgres assumes that 2025-03-01T00:00:00 America/New_York is the same as 2025-03-01T00:00:00-0500 and calculates the offset to that, but that transformation depends on mutable external state (NY state laws) that could change before that time passes.
If you get news of that updated state before March, you now have no way of applying it, as you have thrown away the information of where that seconds since epoch value came from.
You do not confidently know how far away 2025-03-01T00:00:00 America/New_York is from 1970-01-01T00:00:00+0000 until after that time. Even if you decide you're interpreting 1970-01-01T00:00:00+0000 as 1969-12-31T19:00-0500. Postgres assumes that 2025-03-01T00:00:00 America/New_York is the same as 2025-03-01T00:00:00-0500 and calculates the offset to that, but that transformation depends on mutable external state (NY state laws) that could change before that time passes.
If you get news of that updated state before March, you now have no way of applying it, as you have thrown away the information of where that seconds since epoch value came from.