perf: improve prefer-node-protocol's performance

[pralkarz]

Closes #404.

The main change is avoiding an expensive iteration with ReferenceTracker.iterateGlobalReferences by operating directly on ESLint's CallExpressions. I've used a similar approach for ESM imports/exports too rather than sticking with "Program:exit".

Benchmarks and smoke tests were done in the Svelte repository. Both reporting and --fixing works as expected.

master branch:

Rule                                            | Time (ms) | Relative
:-----------------------------------------------|----------:|--------:
n/prefer-node-protocol                          |   902.571 |    51.8%

prefer-node-protocol-performance branch:

Rule                                            | Time (ms) | Relative
:-----------------------------------------------|----------:|--------:
n/prefer-node-protocol                          |    46.429 |     5.2%

[scagood]

Awesome work on both the investigation and the PR for the fix. Thank you!

Everything looks great, but could you revert the Reduce the diff (hopefully) commit please 😅
I think having the functions at the top level is much better!

[pralkarz]

I think having the functions at the top level is much better!

I also prefer that, but I put them back in create since that's what it was like before, and I didn't want to overwhelm you with a bigger diff than necessary. Reverted now though. 😄

[scagood]

🤩 Thank you