Do not activate `test` feature

Hello!

Right now, I believe Rust Analyzer hard codes activating test as a feature. This works well for writing tests, but has the unfortunate side effect that it's difficult to have any cfgs in the rest of a codebase, which should only be active during test.

In my specific usecase, I'm working on an opengl application, and i'd like to make a "HeadlessContext". I use it like this:

#[cfg(test)]
type Gl = HeadlessContext;

#[cfg(not(test))]
type Gl = Context; // the real opengl

Right now, in the above program, Rust Analyzer will infer in all code (outside of test) that I am using HeadlessContext, not Context, which leads to false positive errors (and false negative non-errors).

I don't think the solution to this problem is really that easy though -- if RA simply allowed me to turn off test as a condition, then writing Unit Tests would suddenly get much harder, since i would no longer have access to the RA in such a context. A bit hacky, but a solution where cfg(test) on modules is accepted, but cfg(test) on non-modules is not would fix this usecase, but seems very shaky and perhaps confusing to users.

Thank you for all the hard work on this excellent library! What do you think of the above?

No file or test is named in the issue. Start by locating Rust Analyzer's logic that activates the test feature and read the 15-comment discussion for the intended trade-off between production cfg evaluation and unit-test support. Done means a settled behavior that avoids false diagnostics for non-test code without removing useful test context.