perf: use a dict for linked package store directory membership checks

[jbedard]

The 'not in' check against a list is O(direct deps) for each entry in the transitive npm_package_store_infos of the package, making the linking loop quadratic for packages with many deps.

Changes are visible to end-users:no

Test plan

• Covered by existing test cases

[dzbarsky]

Did you actually measure a win? I feel like linear scan is usually faster than dict for up to 10ish elements so this might be a loss in aggregate

[jbedard]

Did you actually measure a win? I feel like linear scan is usually faster than dict for up to 10ish elements so this might be a loss in aggregate

I can investigate this specific one more and try both.

[jbedard]

I feel like linear scan is usually faster than dict for up to 10ish elements so this might be a loss in aggregate

A robot says:

the dict is win-or-tie everywhere — membership at D=3 (the linear scan's best case) is a tie, and inserts favor the dict ~2x because list.append() is a method call while d[k] = None isn't — so there's no size at which the list wins in aggregate.