fix: normalize resolved result on Windows

[sapphi-red]

#316 seems to break Windows.

If the specifier or the base directory contains a forward slash, the resolved result contained a forward slash.
This causes the same module to be resolved differently (e.g. C:\\foo.js and C:/foo.js) and ends up bundling the same module twice.

For example, import './foo.js'; import '.\\foo.js' should resolve to the same path.

To fix this, without reverting #316, I think it requires thinking about where to normalize the slashes. But it takes some time for me to do that, so in this PR I simply reverted the change.

reverts #316

[sapphi-red]

Suggested change

	path.hash(&mut hasher);
	#[cfg(windows)]
	path.hash(&mut hasher);
	#[cfg(not(windows))]
	path.as_os_str().hash(&mut hasher);

It might be possible to do this. But I'm not sure that's safe for all non-Windows OSs.

[codspeed-hq]

CodSpeed Performance Report

Merging #322 will degrade performances by 17.72%

_{Comparing sapphi-red:fix/normalize-resolved-result-on-windows (8000aeb) with main (290c972)}

Summary

❌ 3 regressions

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	sapphi-red:fix/normalize-resolved-result-on-windows	Change
❌	resolver[multi-thread]	461.7 µs	516.1 µs	-10.54%
❌	resolver[resolve from symlinks]	51.3 ms	62.4 ms	-17.72%
❌	resolver[single-thread]	419.4 µs	482.3 µs	-13.04%

[sapphi-red]

Ah, it seems the test passes with this diff + 08a19d7, but it doesn't with the main branch.

[Boshen]

I wonder whether we should force upstream to normalize all paths ... let me think about this.

Or, change to a simpler, customized hasher.

[sapphi-red]

I thought it would work with a simple revert, but it turns out it needs a bigger change than that 😅.

[sapphi-red]

I'll leave this one for now.

[Boshen]

I'll leave this one for now.

No worries, I'll take over.