Reduce identityFunctionCheck false positives

[Methuselah96]

This PR attempts to reduce the number of false positives identified by identityFunctionCheck by making two changes:

1. Only produce a warning if the result function is only passed one argument.
2. Compare the first argument passed to the result function with the value returned by the result function. There are some edge cases where this could prevent some false positives, although admittedly, they're likely rare (one hypothetical edge case is added as a test).

[netlify]

✅ Deploy Preview for reselect-docs canceled.

Name	Link
🔨 Latest commit	997534f
🔍 Latest deploy log	https://app.netlify.com/sites/reselect-docs/deploys/657642a3ef7d4f0008ba56a5

[codesandbox-ci]

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit 997534f:

Sandbox	Source
Vanilla Typescript	Configuration

[aryaemami59]

This seems like a good idea, the only problem is now something like this won't generate a warning:

const badSelector = createSelector(
  [(state: RootState) => state.todos, (state: RootState) => state.alerts],
  (todos, alerts) => todos
)

I actually thought about doing exactly what you implemented when I was initially working on it, there is definitely a trade off, not sure which one is the way to go.

[Methuselah96]

Agreed, there's no way to make it perfect, but I would definitely lean towards this being a better tradeoff.

1. This still correctly warns for all the practical examples given in Consider dev mode checks for x => x result functions #642.
2. This check has so far produced at least 8 false positives and 0 true positives in my work codebase. Each warning needs to be investigated and takes time to either individually silence, or rework (for each person who runs across them).
3. The cost of false negatives are fairly low. If the check misses more rare forms of the identity function, like the example you provided, it's not that big a deal. We don't have practical examples of people writing identity functions like that in the wild, whereas we do have clear practical examples for a simple identity functions with one arg, and for conditional selectors with multiple args, both of which seem much more common.

All-in-all, it seems clear to me that this provides a better balance for catching true issues, while avoiding false positives.

[aryaemami59]

Yeah, currently I'm working on eslint-plugin for Reselect and RTK, and I think we can go with the solution you provided, and then the other edge cases can be caught using static code analysis. This way we can give accurate warnings without being too strict about it.