There's languages designed with Capabilities in mind. Like, whatever starts the program gets to decide what functionality is exposed to the running program. It's great for situations where you might run untrusted code and want to, as an example, not allow network access, or filesystem access.

More generally there's also sandboxing techniques that runtimes provide. Webassembly for instance is designed for programs to run in their own memory space with a restricted set of functions and, again, Capabilities. This might be nice if you ever work on a cloud application that allows users to upload their own programs and you want to impose limits on those programs. Think AWS Lambda, except the programs running wouldn't necessarily even have access to the filesystem or be able to make web requests unless the user configures that.

It might be a good design space for even more esoteric areas, like device drivers. Like, why worry if your GPU drivers are also collecting telemetry on your computer if you can just turn off that capability?

There's older applications of sandboxing that are a bit further from what you're asking as well; like, iframes on a webpage; allowing code served from different servers you don't necessarily control to run without needing to worry about them reading access tokens from local storage.

Or even BSD Jails and chroot.

Good question 💖

Not a language per se, but subsets of languages used for fantasy consoles usually do not implement import functionality. TIC-80, PICO-8, etc. etc. WIKI I wouldn't call that a feature, but it drives you to write less and more space-optimized code.
Now that I think about it, source code size could be a feature in itself, look at codeGolf-oriented esolangs:

• Pyth
• CJam
• GolfScript
• Microscript/Microscript II
• Seriously
• Rotor
• Minkolang
• Gaia
• Jelly
• 05AB1E
• japt
• and more...

The unnamed language that is compiled by cc.

To elaborate... C[++] is really two different languages, with mostly distinct feature sets, handled in most cases by different compilers, interpreters, parsers, etc.

The unnamed language with keywords like #ifdef and #include which produces text output is a templating system that is functionally independent of the unnamed language with keywords like for and unsigned which actually compiles to a binary.

You can use cpp to run all the logic and conditionals in that first language to produce output, even if you replace the second language with something else like python or assembly.

You can use cc to compile that second language from source to binary, without support from the preprocessor.

That second language, the one that cc understands and compiles, does not have the ability to import functions or values or whatever from other files.

You could check brainfuck, which is designed for extreme minimalism, though not for usefulness (quite the opposite in fact).

I’m not smart enough to know, but what kind of use case(s) do you imagine for this?

Knowing programmers, I think a major unintended side effect of such a paradigm would be HUGE monolithic source files.