chore: refactor windows shim tests

[lukekarrys]

This fixed some flakiness in the windows shim tests. The relevant portion of the fix was to move the t.testdir call to the very first thing in the test. Not sure why this was the case, since tap should be able to handle doing that in multiple tests. From the docs:

While the fixture directory is created synchronously, it is removed asynchronously, because that is the only way to get around ENOTEMPTY errors on Windows.

This means that the next test after one that uses t.testdir() will be deferred until after the end of the current run to completion.

Either way, I ran these tests 3 times and got no flakes. Previously at least one test was flaking every time, so I'm confident that it at least works, even if I'm not 100% sure why.

The rest of the changes were refactoring that I ended up with as I was debugging to see what the issue was.

[npm-cli-bot]

no statistically significant performance changes detected

timing results

app-large	clean	lock-only	cache-only	cache-only peer-deps	modules-only	no-lock	no-cache	no-modules	no-clean	no-clean audit	show-version	run-script
npm@9	40.906 ±0.82	17.139 ±0.03	16.515 ±0.06	19.059 ±1.01	2.659 ±0.06	2.669 ±0.04	2.152 ±0.00	11.282 ±0.03	2.185 ±0.00	3.320 ±0.16	0.364 ±0.00	0.413 ±0.01
#6576	40.690 ±1.26	17.088 ±0.16	16.703 ±0.03	19.129 ±0.40	2.700 ±0.06	2.666 ±0.06	2.217 ±0.04	11.389 ±0.09	2.227 ±0.01	3.368 ±0.01	0.389 ±0.00	0.439 ±0.00

app-medium	clean	lock-only	cache-only	cache-only peer-deps	modules-only	no-lock	no-cache	no-modules	no-clean	no-clean audit	show-version	run-script
npm@9	29.712 ±2.15	13.201 ±0.01	12.676 ±0.17	13.581 ±0.07	2.391 ±0.01	2.401 ±0.03	2.168 ±0.01	8.605 ±0.14	2.041 ±0.03	2.867 ±0.01	0.360 ±0.00	0.411 ±0.00
#6576	27.817 ±0.61	13.459 ±0.12	13.173 ±0.02	14.103 ±0.04	2.443 ±0.00	2.465 ±0.00	2.317 ±0.04	9.049 ±0.10	2.161 ±0.03	2.925 ±0.03	0.392 ±0.00	0.445 ±0.01