But there is? First, by pinning nixpkgs you pin to whatever set of versions was within nixpkgs at the time. Then, if you want to change the version of a package, you can use overlays to poke at what source version is being used to build a package (or what flags it's being built with, etc). There's also nothing preventing your from mixing derivations from multiple nixpkgs pins within a single nixpkgs/nixos fixpoint for your system.
Stuff that in your shell.nix or /etc/nixos/configuration.nix or just nix-build it straight from a small .nix file.
Using an older Nixpkgs checkout might have a higher rate of success but if Nixpkgs never included the version you need in the first place that also doesn't work.
So the mechanisms you describe are in my experience very good in theory but often lacking in practice.
Then you could get the package from a fork of nixpkgs where it is present.
Yeah, but also keep in mind that uses the older version, plus older versions of all its transitive dependencies (from the same nixpkgs checkout). So, you could end up with multiple versions of glibc and other commonly-used libraries, but worse -- many known security vulnerabilities as well.
Another practical annoyance is that overriding attributes often does not work across a derivation. E.g. overriding the version attribute does not override the same version in the src attribute (which is used to download the sources) [1]. So, you may have to override the attribute in several places.
[1] Unless the derivation is defined in a non-canonical way: https://discourse.nixos.org/t/avoid-rec-expresions-in-nixpkg...
When the version matters, Nix has separate packages for separate versions. There are separate Nix packages for Python 2.7, 3.7, 3.8, 3.9 or 3.10, and there's OpenJDK 8, 11 and 17, etc.
You probably don't want to depend on an overly specific version anyway. If a security fix for one of your dependencies is released, you want it. Furthermore, CI ensures that everything within a channel is maximally consistent.